CLandmarksMap.cpp

Go to the documentation of this file.

/* +------------------------------------------------------------------------+
   |                     Mobile Robot Programming Toolkit (MRPT)            |
   |                          https://www.mrpt.org/                         |
   |                                                                        |
   | Copyright (c) 2005-2019, Individual contributors, see AUTHORS file     |
   | See: https://www.mrpt.org/Authors - All rights reserved.               |
   | Released under BSD License. See: https://www.mrpt.org/License          |
   +------------------------------------------------------------------------+ */

#include "vision-precomp.h"  // Precompiled headers

#include <mrpt/io/CFileOutputStream.h>
#include <mrpt/maps/CLandmark.h>
#include <mrpt/maps/CLandmarksMap.h>
#include <mrpt/math/geometry.h>
#include <mrpt/math/ops_matrices.h>
#include <mrpt/obs/CObservation2DRangeScan.h>
#include <mrpt/obs/CObservationBeaconRanges.h>
#include <mrpt/obs/CObservationGPS.h>
#include <mrpt/obs/CObservationImage.h>
#include <mrpt/obs/CObservationRobotPose.h>
#include <mrpt/obs/CObservationStereoImages.h>
#include <mrpt/obs/CObservationVisualLandmarks.h>
#include <mrpt/opengl/CEllipsoid.h>
#include <mrpt/opengl/CGridPlaneXY.h>
#include <mrpt/opengl/COpenGLScene.h>
#include <mrpt/poses/CPointPDFGaussian.h>
#include <mrpt/random.h>
#include <mrpt/system/os.h>
#include <Eigen/Dense>

using namespace mrpt;
using namespace mrpt::math;
using namespace mrpt::maps;
using namespace mrpt::obs;
using namespace mrpt::poses;
using namespace mrpt::tfest;
using namespace mrpt::random;
using namespace mrpt::system;
using namespace mrpt::vision;
using namespace mrpt::img;
using namespace mrpt::containers;
using namespace std;
using mrpt::maps::internal::TSequenceLandmarks;

//  =========== Begin of Map definition ============
MAP_DEFINITION_REGISTER("CLandmarksMap,landmarksMap", mrpt::maps::CLandmarksMap)

CLandmarksMap::TMapDefinition::TMapDefinition() = default;
void CLandmarksMap::TMapDefinition::loadFromConfigFile_map_specific(
    const mrpt::config::CConfigFileBase& source,
    const std::string& sectionNamePrefix)
{
    // [<sectionNamePrefix>+"_creationOpts"]
    const std::string sSectCreation =
        sectionNamePrefix + string("_creationOpts");
    this->initialBeacons.clear();
    const unsigned int nBeacons = source.read_int(sSectCreation, "nBeacons", 0);
    for (unsigned int q = 1; q <= nBeacons; q++)
    {
        TPairIdBeacon newPair;
        newPair.second =
            source.read_int(sSectCreation, format("beacon_%03u_ID", q), 0);

        newPair.first.x =
            source.read_float(sSectCreation, format("beacon_%03u_x", q), 0);
        newPair.first.y =
            source.read_float(sSectCreation, format("beacon_%03u_y", q), 0);
        newPair.first.z =
            source.read_float(sSectCreation, format("beacon_%03u_z", q), 0);

        this->initialBeacons.push_back(newPair);
    }

    insertionOpts.loadFromConfigFile(
        source, sectionNamePrefix + string("_insertOpts"));
    likelihoodOpts.loadFromConfigFile(
        source, sectionNamePrefix + string("_likelihoodOpts"));
}

void CLandmarksMap::TMapDefinition::dumpToTextStream_map_specific(
    std::ostream& out) const
{
    out << mrpt::format(
        "number of initial beacons                = %u\n",
        (int)initialBeacons.size());

    out << mrpt::format("      ID         (X,Y,Z)\n");
    out << mrpt::format(
        "--------------------------------------------------------\n");
    for (const auto& initialBeacon : initialBeacons)
        out << mrpt::format(
            "      %03u         (%8.03f,%8.03f,%8.03f)\n", initialBeacon.second,
            initialBeacon.first.x, initialBeacon.first.y,
            initialBeacon.first.z);

    this->insertionOpts.dumpToTextStream(out);
    this->likelihoodOpts.dumpToTextStream(out);
}

mrpt::maps::CMetricMap* CLandmarksMap::internal_CreateFromMapDefinition(
    const mrpt::maps::TMetricMapInitializer& _def)
{
    const CLandmarksMap::TMapDefinition& def =
        *dynamic_cast<const CLandmarksMap::TMapDefinition*>(&_def);
    auto* obj = new CLandmarksMap();

    for (const auto& initialBeacon : def.initialBeacons)
    {
        CLandmark lm;

        lm.createOneFeature();
        lm.features[0].type = featBeacon;

        lm.features[0].keypoint.ID = initialBeacon.second;
        lm.ID = initialBeacon.second;

        lm.pose_mean = initialBeacon.first;

        lm.pose_cov_11 = lm.pose_cov_22 = lm.pose_cov_33 = lm.pose_cov_12 =
            lm.pose_cov_13 = lm.pose_cov_23 = square(0.01f);

        obj->landmarks.push_back(lm);
    }

    obj->insertionOptions = def.insertionOpts;
    obj->likelihoodOptions = def.likelihoodOpts;
    return obj;
}
//  =========== End of Map definition Block =========

IMPLEMENTS_SERIALIZABLE(CLandmarksMap, CMetricMap, mrpt::maps)

/*---------------------------------------------------------------
                Static variables initialization
  ---------------------------------------------------------------*/
std::map<
    std::pair<
        mrpt::maps::CLandmark::TLandmarkID, mrpt::maps::CLandmark::TLandmarkID>,
    double>
    CLandmarksMap::_mEDD;
mrpt::maps::CLandmark::TLandmarkID CLandmarksMap::_mapMaxID;
bool CLandmarksMap::_maxIDUpdated = false;

void CLandmarksMap::internal_clear() { landmarks.clear(); }

size_t CLandmarksMap::size() const { return landmarks.size(); }

uint8_t CLandmarksMap::serializeGetVersion() const { return 0; }
void CLandmarksMap::serializeTo(mrpt::serialization::CArchive& out) const
{
    uint32_t n = landmarks.size();

    // First, write the number of landmarks:
    out << n;

    // Write all landmarks:
    for (const auto& landmark : landmarks) out << landmark;
}

void CLandmarksMap::serializeFrom(
    mrpt::serialization::CArchive& in, uint8_t version)
{
    switch (version)
    {
        case 0:
        {
            uint32_t n, i;
            CLandmark lm;

            // Delete previous content of map:
            // -------------------------------------
            landmarks.clear();

            // Load from stream:
            // -------------------------------------
            in >> n;

            landmarks.clear();  // resize(n);

            // Read all landmarks:
            for (i = 0; i < n; i++)
            {
                in >> lm;
                landmarks.push_back(lm);
            }
        }
        break;
        default:
            MRPT_THROW_UNKNOWN_SERIALIZATION_VERSION(version);
    };
}

/*---------------------------------------------------------------
                    computeObservationLikelihood
  ---------------------------------------------------------------*/
double CLandmarksMap::internal_computeObservationLikelihood(
    const CObservation& obs, const CPose3D& robotPose3D)
{
    MRPT_START

    if (CLASS_ID(CObservation2DRangeScan) == obs.GetRuntimeClass() &&
        insertionOptions.insert_Landmarks_from_range_scans)
    {
        /********************************************************************
                        OBSERVATION TYPE: CObservation2DRangeScan
            ********************************************************************/
        const auto& o = static_cast<const CObservation2DRangeScan&>(obs);
        CLandmarksMap auxMap;
        CPose2D sensorPose2D(robotPose3D + o.sensorPose);

        auxMap.loadOccupancyFeaturesFrom2DRangeScan(
            o, &robotPose3D, likelihoodOptions.rangeScan2D_decimation);

        // And compute its likelihood:
        return computeLikelihood_RSLC_2007(&auxMap, sensorPose2D);
    }  // end of likelihood of 2D range scan:
    else if (CLASS_ID(CObservationStereoImages) == obs.GetRuntimeClass())
    {
        /********************************************************************
                        OBSERVATION TYPE: CObservationStereoImages
                Lik. between "this" and "auxMap";
            ********************************************************************/
        const auto& o = static_cast<const CObservationStereoImages&>(obs);

        CLandmarksMap auxMap;
        auxMap.insertionOptions = insertionOptions;
        auxMap.loadSiftFeaturesFromStereoImageObservation(
            o, CLandmarksMap::_mapMaxID, likelihoodOptions.SIFT_feat_options);
        auxMap.changeCoordinatesReference(robotPose3D);

        // ACCESS TO STATIC VARIABLE
        // std::cout << "_mapMaxID, from " << CLandmarksMap::_mapMaxID << " to
        // ";
        if (!CLandmarksMap::_maxIDUpdated)
        {
            CLandmarksMap::_mapMaxID += auxMap.size();
            CLandmarksMap::_maxIDUpdated = true;
        }  // end if

        // std::cout << CLandmarksMap::_mapMaxID <<std::endl;
        return computeLikelihood_SIFT_LandmarkMap(&auxMap);

    }  // end of likelihood of Stereo Images scan:
    else if (CLASS_ID(CObservationBeaconRanges) == obs.GetRuntimeClass())
    {
        /********************************************************************

                        OBSERVATION TYPE: CObservationBeaconRanges

                Lik. between "this" and "auxMap";

            ********************************************************************/
        const auto& o = static_cast<const CObservationBeaconRanges&>(obs);

        std::deque<CObservationBeaconRanges::TMeasurement>::const_iterator it;
        TSequenceLandmarks::iterator lm_it;
        CPointPDFGaussian beaconPDF;
        CPoint3D point3D, beacon3D;
        double ret = 0;  // 300;

        for (it = o.sensedData.begin(); it != o.sensedData.end(); it++)
        {
            // Look for the beacon in this map:
            unsigned int sensedID = it->beaconID;
            bool found = false;

            for (lm_it = landmarks.begin(); !found && lm_it != landmarks.end();
                 lm_it++)
            {
                if ((lm_it->getType() == featBeacon) &&
                    (lm_it->ID == sensedID) &&
                    (!std::isnan(it->sensedDistance)))
                {
                    lm_it->getPose(beaconPDF);
                    beacon3D = beaconPDF.mean;

                    // Compute the 3D position of the sensor:
                    point3D = robotPose3D + it->sensorLocationOnRobot;

                    const float expectedRange = point3D.distanceTo(beacon3D);
                    float sensedDist = it->sensedDistance;
                    if (sensedDist < 0) sensedDist = 0;

                    float sensorStd =
                        likelihoodOptions.beaconRangesUseObservationStd
                            ? o.stdError
                            : likelihoodOptions.beaconRangesStd;
                    ret +=
                        (-0.5f *
                         square((expectedRange - sensedDist) / sensorStd));
                    found = true;
                }
            }

            // If not found, uniform distribution:
            if (!found)
            {
                if (o.maxSensorDistance != o.minSensorDistance)
                    ret +=
                        log(1.0 / (o.maxSensorDistance - o.minSensorDistance));
            }

        }  // for each sensed beacon "it"

        MRPT_CHECK_NORMAL_NUMBER(ret);
        return ret;

    }  // end of likelihood of CObservationBeaconRanges
    else if (CLASS_ID(CObservationRobotPose) == obs.GetRuntimeClass())
    {
        /********************************************************************

                OBSERVATION TYPE: CObservationRobotPose

                Lik. between "this" and "robotPose";

        ********************************************************************/
        const auto& o = static_cast<const CObservationRobotPose&>(obs);

        // Compute the 3D position of the sensor:
        CPose3D sensorPose3D = robotPose3D + o.sensorPose;

        // Compute the likelihood according to mahalanobis distance between
        // poses:
        CMatrixD dij(1, 6), Cij(6, 6), Cij_1;
        dij(0, 0) = o.pose.mean.x() - sensorPose3D.x();
        dij(0, 1) = o.pose.mean.y() - sensorPose3D.y();
        dij(0, 2) = o.pose.mean.z() - sensorPose3D.z();
        dij(0, 3) = wrapToPi(o.pose.mean.yaw() - sensorPose3D.yaw());
        dij(0, 4) = wrapToPi(o.pose.mean.pitch() - sensorPose3D.pitch());
        dij(0, 5) = wrapToPi(o.pose.mean.roll() - sensorPose3D.roll());

        // Equivalent covariance from "i" to "j":
        Cij = CMatrixDouble(o.pose.cov);
        Cij_1 = Cij.inverse_LLt();

        double distMahaFlik2 = mrpt::math::multiply_HCHt_scalar(dij, Cij_1);
        double ret =
            -0.5 * (distMahaFlik2 / square(likelihoodOptions.extRobotPoseStd));

        MRPT_CHECK_NORMAL_NUMBER(ret);
        return ret;

    }  // end of likelihood of CObservation
    else if (CLASS_ID(CObservationGPS) == obs.GetRuntimeClass())
    {
        /********************************************************************

                        OBSERVATION TYPE: CObservationGPS

        ********************************************************************/
        const auto& o = static_cast<const CObservationGPS&>(obs);
        // Compute the 3D position of the sensor:
        CPoint3D point3D = CPoint3D(robotPose3D);
        CPoint3D GPSpose;
        double x, y;
        double earth_radius = 6378137;

        if ((o.has_GGA_datum) &&
            (likelihoodOptions.GPSOrigin.min_sat <=
             o.getMsgByClass<gnss::Message_NMEA_GGA>().fields.satellitesUsed))
        {
            // Compose GPS robot position

            x = DEG2RAD(
                    (o.getMsgByClass<gnss::Message_NMEA_GGA>()
                         .fields.longitude_degrees -
                     likelihoodOptions.GPSOrigin.longitude)) *
                earth_radius * 1.03;
            y = DEG2RAD(
                    (o.getMsgByClass<gnss::Message_NMEA_GGA>()
                         .fields.latitude_degrees -
                     likelihoodOptions.GPSOrigin.latitude)) *
                earth_radius * 1.15;
            GPSpose.x(
                (x * cos(likelihoodOptions.GPSOrigin.ang) +
                 y * sin(likelihoodOptions.GPSOrigin.ang) +
                 likelihoodOptions.GPSOrigin.x_shift));
            GPSpose.y(
                (-x * sin(likelihoodOptions.GPSOrigin.ang) +
                 y * cos(likelihoodOptions.GPSOrigin.ang) +
                 likelihoodOptions.GPSOrigin.y_shift));
            GPSpose.z(
                (o.getMsgByClass<gnss::Message_NMEA_GGA>()
                     .fields.altitude_meters -
                 likelihoodOptions.GPSOrigin.altitude));
            // std::cout<<"GPSpose calculo: "<<GPSpose.x<<","<<GPSpose.y<<"\n";

            //-------------------------------//
            // sigmaGPS =
            // f(o.getMsgByClass<gnss::Message_NMEA_GGA>().fields.satellitesUsed)
            // //funcion del numero de satelites
            //-------------------------------//

            // std::cout<<"datos de longitud y latitud:
            // "<<o.getMsgByClass<gnss::Message_NMEA_GGA>().fields.longitude_degrees<<","<<o.getMsgByClass<gnss::Message_NMEA_GGA>().fields.latitude_degrees<<","<<"\n";
            // std::cout<<"x,y sin rotar: "<<x<<","<<y<<","<<"\n";
            // std::cout<<"angulo: "<<likelihoodOptions.GPSOrigin.ang<<"\n";
            // std::cout<<"desp x,y:
            // "<<likelihoodOptions.GPSOrigin.x_shift<<","<<likelihoodOptions.GPSOrigin.y_shift<<"\n";
            // std::cout<<"GPS ORIGIN    :
            // "<<likelihoodOptions.GPSOrigin.longitude<<","<<likelihoodOptions.GPSOrigin.latitude<<","<<likelihoodOptions.GPSOrigin.altitude<<"\n";
            // std::cout<<"POSE DEL ROBOT:
            // "<<point3D.x<<","<<point3D.y<<","<<point3D.z<<"\n";
            // std::cout<<"POSE GPS      :
            // "<<GPSpose.x<<","<<GPSpose.y<<","<<GPSpose.z<<"\n";
            // std::cin.get();

            float distance = GPSpose.distanceTo(point3D);

            // std::cout<<"likel gps:"<<-0.5f*square( ( distance
            // )/likelihoodOptions.GPS_sigma)<<"\n";;
            double ret =
                -0.5f * square((distance) / likelihoodOptions.GPS_sigma);
            MRPT_CHECK_NORMAL_NUMBER(ret);
            return ret;
        }
        else
            return 0.5;
    }  // end of likelihood of CObservationGPS
    else
    {
        /********************************************************************

                    OBSERVATION TYPE: Unknown

        ********************************************************************/
        return 0.5;
    }

    MRPT_END
}

/*---------------------------------------------------------------
                        insertObservation
  ---------------------------------------------------------------*/
bool CLandmarksMap::internal_insertObservation(
    const CObservation& obs, const CPose3D* robotPose)
{
    MRPT_START

    CPose2D robotPose2D;
    CPose3D robotPose3D;

    if (robotPose)
    {
        robotPose2D = CPose2D(*robotPose);
        robotPose3D = (*robotPose);
    }
    else
    {
        // Default values are (0,0,0)
    }

    if (CLASS_ID(CObservationImage) == obs.GetRuntimeClass() &&
        insertionOptions.insert_SIFTs_from_monocular_images)
    {
        /********************************************************************

                        OBSERVATION TYPE: CObservationImage

            ********************************************************************/
        const auto& o = static_cast<const CObservationImage&>(obs);
        CLandmarksMap tempMap;

        // 1) Load the features in a temporary 3D landmarks map:
        tempMap.loadSiftFeaturesFromImageObservation(
            o, insertionOptions.SIFT_feat_options);

        // 2) This temp. map must be moved to its real position on the global
        // reference coordinates:
        tempMap.changeCoordinatesReference(robotPose3D);

        // 3) Fuse that map with the current contents of "this" one:
        fuseWith(tempMap);

        // DONE!!

        // Observation was successfully inserted into the map
        // --------------------------------------------------------
        return true;
    }
    //  else
    //  if ( CLASS_ID(CObservation2DRangeScan)==obs.GetRuntimeClass() &&
    //        insertionOptions.insert_Landmarks_from_range_scans)
    //  {
    /********************************************************************

                    OBSERVATION TYPE: CObservation2DRangeScan

        ********************************************************************/
    /*      CObservation2DRangeScan *o = (CObservation2DRangeScan*) obs;
            CLandmarksMap           tempMap;

            // Load into the map:
            tempMap.loadOccupancyFeaturesFrom2DRangeScan(*o, robotPose);
            fuseWith( tempMap );

            // Observation was successfully inserted into the map
            // --------------------------------------------------------
            return true;
        }                       */
    else if (
        CLASS_ID(CObservationStereoImages) == obs.GetRuntimeClass() &&
        insertionOptions.insert_SIFTs_from_stereo_images)
    {
        /********************************************************************
                        OBSERVATION TYPE: CObservationStereoImages
            ********************************************************************/
        const auto& o = static_cast<const CObservationStereoImages&>(obs);

        // Change coordinates ref:
        CLandmarksMap auxMap;
        auxMap.insertionOptions = insertionOptions;
        auxMap.loadSiftFeaturesFromStereoImageObservation(
            o, CLandmarksMap::_mapMaxID, insertionOptions.SIFT_feat_options);
        auxMap.changeCoordinatesReference(robotPose3D);

        fuseWith(auxMap);

        // Observation was successfully inserted into the map
        // --------------------------------------------------------
        return true;
    }
    else if (CLASS_ID(CObservationVisualLandmarks) == obs.GetRuntimeClass())
    {
        /********************************************************************

                        OBSERVATION TYPE:  CObservationVisualLandmarks

            ********************************************************************/
        const auto& o = static_cast<const CObservationVisualLandmarks&>(obs);

        // Change coordinates ref:
        CLandmarksMap auxMap;
        CPose3D acumTransform(robotPose3D + o.refCameraPose);
        auxMap.changeCoordinatesReference(acumTransform, &o.landmarks);

        // Fuse with current:
        fuseWith(auxMap, true);

        // Observation was successfully inserted into the map
        // --------------------------------------------------------
        return true;
    }
    else
    {
        /********************************************************************
                    OBSERVATION TYPE: Unknown
        ********************************************************************/
        return false;
    }

    MRPT_END
}

/*---------------------------------------------------------------
                computeMatchingWith2D
  ---------------------------------------------------------------*/
void CLandmarksMap::computeMatchingWith2D(
    const mrpt::maps::CMetricMap* otherMap, const CPose2D& otherMapPose,
    float maxDistForCorrespondence, float maxAngularDistForCorrespondence,
    const CPose2D& angularDistPivotPoint, TMatchingPairList& correspondences,
    float& correspondencesRatio, float* sumSqrDist, bool onlyKeepTheClosest,
    bool onlyUniqueRobust) const
{
    MRPT_UNUSED_PARAM(onlyKeepTheClosest);
    MRPT_UNUSED_PARAM(onlyUniqueRobust);
    MRPT_UNUSED_PARAM(sumSqrDist);
    MRPT_UNUSED_PARAM(angularDistPivotPoint);
    MRPT_UNUSED_PARAM(maxDistForCorrespondence);
    MRPT_UNUSED_PARAM(maxAngularDistForCorrespondence);

    MRPT_START

    CLandmarksMap auxMap;
    CPose3D otherMapPose3D(otherMapPose);

    correspondencesRatio = 0;

    // Check the other map class:
    ASSERT_(otherMap->GetRuntimeClass() == CLASS_ID(CLandmarksMap));
    const auto* otherMap2 = static_cast<const CLandmarksMap*>(otherMap);
    std::vector<bool> otherCorrespondences;

    // Coordinates change:
    auxMap.changeCoordinatesReference(otherMapPose3D, otherMap2);

    //// Use the 3D matching method:
    computeMatchingWith3DLandmarks(
        otherMap2, correspondences, correspondencesRatio, otherCorrespondences);

    MRPT_END
}

/*---------------------------------------------------------------
                loadSiftFeaturesFromImageObservation
  ---------------------------------------------------------------*/
void CLandmarksMap::loadSiftFeaturesFromImageObservation(
    const CObservationImage& obs,
    const mrpt::vision::CFeatureExtraction::TOptions& feat_options)
{
    CImage corImg;
    CPointPDFGaussian landmark3DPositionPDF;
    float d = insertionOptions.SIFTsLoadDistanceOfTheMean;
    float width = insertionOptions.SIFTsLoadEllipsoidWidth;
    CMatrixDouble33 P, D;
    CLandmark lm;

    vision::CFeatureExtraction fExt;
    vision::CFeatureList siftList;  // vision::TSIFTFeatureList siftList;
    vision::CFeatureList::iterator
        sift;  // vision::TSIFTFeatureList::iterator    sift;

    // Timestamp:
    lm.timestampLastSeen = obs.timestamp;
    lm.seenTimesCount = 1;

    // Remove distortion:
    corImg =
        obs.image;  // obs.image.correctDistortion(obs.intrinsicParams,obs.distortionParams);

    // Extract SIFT features:
    fExt.options = feat_options;
    fExt.detectFeatures(
        corImg, siftList);  // vision::computeSiftFeatures(corImg, siftList );

    // Save them as 3D landmarks:
    landmarks.clear();  // resize( siftList.size() );

    for (sift = siftList.begin(); sift != siftList.end(); sift++)
    {
        // Find the 3D position from the pixels
        //  coordinates and the camera intrinsic matrix:
        mrpt::math::TPoint3D dir = vision::pixelTo3D(
            sift->keypoint.pt, obs.cameraParams.intrinsicParams);

        // Compute the mean and covariance of the landmark gaussian 3D position,
        //  from the unitary direction vector and a given distance:
        // --------------------------------------------------------------------------
        landmark3DPositionPDF.mean = CPoint3D(dir);  // The mean is easy :-)
        landmark3DPositionPDF.mean *= d;

        // The covariance:
        P = math::generateAxisBaseFromDirection(dir.x, dir.y, dir.z);

        // Diagonal matrix (with the "size" of the ellipsoid)
        D(0, 0) = square(0.5 * d);
        D(1, 1) = square(width);
        D(2, 2) = square(width);

        // Finally, compute the covariance!
        landmark3DPositionPDF.cov = mrpt::math::multiply_HCHt(P, D);

        // Save into the landmarks vector:
        // --------------------------------------------
        lm.features.resize(1);
        lm.features[0] = (*sift);

        CPoint3D Normal = landmark3DPositionPDF.mean;
        Normal *= -1 / Normal.norm();

        lm.normal = Normal.asTPoint();

        lm.pose_mean = landmark3DPositionPDF.mean.asTPoint();

        lm.pose_cov_11 = landmark3DPositionPDF.cov(0, 0);
        lm.pose_cov_22 = landmark3DPositionPDF.cov(1, 1);
        lm.pose_cov_33 = landmark3DPositionPDF.cov(2, 2);
        lm.pose_cov_12 = landmark3DPositionPDF.cov(0, 1);
        lm.pose_cov_13 = landmark3DPositionPDF.cov(0, 2);
        lm.pose_cov_23 = landmark3DPositionPDF.cov(1, 2);

        landmarks.push_back(lm);
    }

}  // end loadSiftFeaturesFromImageObservation

/*---------------------------------------------------------------
                loadSiftFeaturesFromStereoImagesObservation
  ---------------------------------------------------------------*/
void CLandmarksMap::loadSiftFeaturesFromStereoImageObservation(
    const CObservationStereoImages& obs, mrpt::maps::CLandmark::TLandmarkID fID,
    const mrpt::vision::CFeatureExtraction::TOptions& feat_options)
{
    MRPT_START

    vision::CFeatureExtraction fExt;
    vision::CFeatureList leftSiftList, rightSiftList;
    vision::CMatchedFeatureList matchesList;
    vision::TMatchingOptions matchingOptions;
    vision::TStereoSystemParams stereoParams;

    CLandmarksMap landmarkMap;

    // Extract SIFT features:
    fExt.options = feat_options;  // OLD:
    // fExt.options.SIFTOptions.implementation =
    // vision::CFeatureExtraction::Hess;

    // Default: Hess implementation
    fExt.detectFeatures(
        obs.imageLeft, leftSiftList, fID,
        insertionOptions.SIFTs_numberOfKLTKeypoints);
    fExt.detectFeatures(
        obs.imageRight, rightSiftList, fID,
        insertionOptions.SIFTs_numberOfKLTKeypoints);

    matchingOptions.matching_method =
        vision::TMatchingOptions::mmDescriptorSIFT;
    matchingOptions.epipolar_TH = insertionOptions.SIFTs_epipolar_TH;
    matchingOptions.EDD_RATIO = insertionOptions.SiftCorrRatioThreshold;
    matchingOptions.maxEDD_TH = insertionOptions.SiftEDDThreshold;
    vision::matchFeatures(
        leftSiftList, rightSiftList, matchesList, matchingOptions);

    if (insertionOptions.PLOT_IMAGES)
    {
        std::cerr << "Warning: insertionOptions.PLOT_IMAGES has no effect "
                     "since MRPT 0.9.1\n";
    }

    // obs.imageLeft.saveToFile("LImage.jpg");
    // obs.imageRight.saveToFile("RImage.jpg");
    // FILE *fmt;
    // fmt = os::fopen( "matchesRBPF.txt", "at" );
    // os::fprintf( fmt, "%d\n", (unsigned int)matchesList.size() );
    // os::fclose(fmt);
    // matchesList.saveToTextFile("matches.txt");

    // Feature Projection to 3D
    // Parameters of the stereo rig

    stereoParams.K = obs.leftCamera.intrinsicParams;
    stereoParams.stdPixel = insertionOptions.SIFTs_stdXY;
    stereoParams.stdDisp = insertionOptions.SIFTs_stdDisparity;
    stereoParams.baseline = obs.rightCameraPose.x();
    stereoParams.minZ = 0.0f;
    stereoParams.maxZ = insertionOptions.SIFTs_stereo_maxDepth;

    size_t numM = matchesList.size();
    vision::projectMatchedFeatures(matchesList, stereoParams, *this);

    // Add Timestamp and the "Seen-Times" counter
    TCustomSequenceLandmarks::iterator ii;
    for (ii = landmarks.begin(); ii != landmarks.end(); ii++)
    {
        (*ii).timestampLastSeen = obs.timestamp;
        (*ii).seenTimesCount = 1;
    }

    printf(
        "%u (out of %u) corrs!\n", static_cast<unsigned>(landmarks.size()),
        static_cast<unsigned>(numM));

    // CLandmarksMap::_maxMapID = fID;

    // Project landmarks according to the ref. camera pose:
    changeCoordinatesReference(mrpt::poses::CPose3D(obs.cameraPose));

    MRPT_END
}

/*---------------------------------------------------------------
                changeCoordinatesReference
  ---------------------------------------------------------------*/
void CLandmarksMap::changeCoordinatesReference(const CPose3D& newOrg)
{
    TSequenceLandmarks::iterator lm;

    CMatrixDouble44 HM;
    newOrg.getHomogeneousMatrix(HM);

    // Build the rotation only transformation:
    double R11 = HM(0, 0);
    double R12 = HM(0, 1);
    double R13 = HM(0, 2);
    double R21 = HM(1, 0);
    double R22 = HM(1, 1);
    double R23 = HM(1, 2);
    double R31 = HM(2, 0);
    double R32 = HM(2, 1);
    double R33 = HM(2, 2);

    double c11, c22, c33, c12, c13, c23;

    // Change the reference of each individual landmark:
    // ----------------------------------------------------
    for (lm = landmarks.begin(); lm != landmarks.end(); lm++)
    {
        // Transform the pose mean & covariance:
        // ---------------------------------------------------------
        newOrg.composePoint(lm->pose_mean, lm->pose_mean);

        float C11 = lm->pose_cov_11;
        float C22 = lm->pose_cov_22;
        float C33 = lm->pose_cov_33;
        float C12 = lm->pose_cov_12;
        float C13 = lm->pose_cov_13;
        float C23 = lm->pose_cov_23;

        // The covariance:  cov = M * cov * (~M);
        c11 = R11 * (C11 * R11 + C12 * R12 + C13 * R13) +
              R12 * (C12 * R11 + C22 * R12 + C23 * R13) +
              R13 * (C13 * R11 + C23 * R12 + C33 * R13);
        c12 = (C11 * R11 + C12 * R12 + C13 * R13) * R21 +
              (C12 * R11 + C22 * R12 + C23 * R13) * R22 +
              (C13 * R11 + C23 * R12 + C33 * R13) * R23;
        c13 = (C11 * R11 + C12 * R12 + C13 * R13) * R31 +
              (C12 * R11 + C22 * R12 + C23 * R13) * R32 +
              (C13 * R11 + C23 * R12 + C33 * R13) * R33;
        c22 = R21 * (C11 * R21 + C12 * R22 + C13 * R23) +
              R22 * (C12 * R21 + C22 * R22 + C23 * R23) +
              R23 * (C13 * R21 + C23 * R22 + C33 * R23);
        c23 = (C11 * R21 + C12 * R22 + C13 * R23) * R31 +
              (C12 * R21 + C22 * R22 + C23 * R23) * R32 +
              (C13 * R21 + C23 * R22 + C33 * R23) * R33;
        c33 = R31 * (C11 * R31 + C12 * R32 + C13 * R33) +
              R32 * (C12 * R31 + C22 * R32 + C23 * R33) +
              R33 * (C13 * R31 + C23 * R32 + C33 * R33);

        // save into the landmark:
        lm->pose_cov_11 = c11;
        lm->pose_cov_22 = c22;
        lm->pose_cov_33 = c33;
        lm->pose_cov_12 = c12;
        lm->pose_cov_13 = c13;
        lm->pose_cov_23 = c23;

        // Rotate the normal:         lm->normal = rot + lm->normal;
        // ---------------------------------------------------------
        float Nx = lm->normal.x;
        float Ny = lm->normal.y;
        float Nz = lm->normal.z;

        lm->normal.x = Nx * R11 + Ny * R12 + Nz * R13;
        lm->normal.y = Nx * R21 + Ny * R22 + Nz * R23;
        lm->normal.z = Nx * R31 + Ny * R32 + Nz * R33;
    }

    // For updating the KD-Tree
    landmarks.hasBeenModifiedAll();
}

/*---------------------------------------------------------------
                changeCoordinatesReference
  ---------------------------------------------------------------*/
void CLandmarksMap::changeCoordinatesReference(
    const CPose3D& newOrg, const mrpt::maps::CLandmarksMap* otherMap)
{
    TSequenceLandmarks::const_iterator lm;
    CLandmark newLandmark;

    CMatrixDouble44 HM;
    newOrg.getHomogeneousMatrix(HM);

    // Build the rotation only transformation:
    double R11 = HM(0, 0);
    double R12 = HM(0, 1);
    double R13 = HM(0, 2);
    double R21 = HM(1, 0);
    double R22 = HM(1, 1);
    double R23 = HM(1, 2);
    double R31 = HM(2, 0);
    double R32 = HM(2, 1);
    double R33 = HM(2, 2);

    double c11, c22, c33, c12, c13, c23;

    landmarks.clear();

    // Change the reference of each individual landmark:
    // ----------------------------------------------------
    for (lm = otherMap->landmarks.begin(); lm != otherMap->landmarks.end();
         lm++)
    {
        // Transform the pose mean & covariance:
        // ---------------------------------------------------------
        // The mean:     mean = newReferenceBase + mean;
        newOrg.composePoint(lm->pose_mean, newLandmark.pose_mean);

        float C11 = lm->pose_cov_11;
        float C22 = lm->pose_cov_22;
        float C33 = lm->pose_cov_33;
        float C12 = lm->pose_cov_12;
        float C13 = lm->pose_cov_13;
        float C23 = lm->pose_cov_23;

        // The covariance:  cov = M * cov * (~M);
        c11 = R11 * (C11 * R11 + C12 * R12 + C13 * R13) +
              R12 * (C12 * R11 + C22 * R12 + C23 * R13) +
              R13 * (C13 * R11 + C23 * R12 + C33 * R13);
        c12 = (C11 * R11 + C12 * R12 + C13 * R13) * R21 +
              (C12 * R11 + C22 * R12 + C23 * R13) * R22 +
              (C13 * R11 + C23 * R12 + C33 * R13) * R23;
        c13 = (C11 * R11 + C12 * R12 + C13 * R13) * R31 +
              (C12 * R11 + C22 * R12 + C23 * R13) * R32 +
              (C13 * R11 + C23 * R12 + C33 * R13) * R33;
        c22 = R21 * (C11 * R21 + C12 * R22 + C13 * R23) +
              R22 * (C12 * R21 + C22 * R22 + C23 * R23) +
              R23 * (C13 * R21 + C23 * R22 + C33 * R23);
        c23 = (C11 * R21 + C12 * R22 + C13 * R23) * R31 +
              (C12 * R21 + C22 * R22 + C23 * R23) * R32 +
              (C13 * R21 + C23 * R22 + C33 * R23) * R33;
        c33 = R31 * (C11 * R31 + C12 * R32 + C13 * R33) +
              R32 * (C12 * R31 + C22 * R32 + C23 * R33) +
              R33 * (C13 * R31 + C23 * R32 + C33 * R33);

        // save into the landmark:
        newLandmark.pose_cov_11 = c11;
        newLandmark.pose_cov_22 = c22;
        newLandmark.pose_cov_33 = c33;
        newLandmark.pose_cov_12 = c12;
        newLandmark.pose_cov_13 = c13;
        newLandmark.pose_cov_23 = c23;

        // Rotate the normal:         lm->normal = rot + lm->normal;
        // ---------------------------------------------------------
        float Nx = lm->normal.x;
        float Ny = lm->normal.y;
        float Nz = lm->normal.z;

        newLandmark.normal.x = Nx * R11 + Ny * R12 + Nz * R13;
        newLandmark.normal.y = Nx * R21 + Ny * R22 + Nz * R23;
        newLandmark.normal.z = Nx * R31 + Ny * R32 + Nz * R33;

        newLandmark.ID = lm->ID;

        newLandmark.features = lm->features;

        newLandmark.timestampLastSeen = lm->timestampLastSeen;
        newLandmark.seenTimesCount = lm->seenTimesCount;

        landmarks.push_back(newLandmark);
    }
}

/*---------------------------------------------------------------
                        fuseWith
  ---------------------------------------------------------------*/
void CLandmarksMap::fuseWith(CLandmarksMap& other, bool justInsertAllOfThem)
{
    MRPT_START

    // std::cout << "Entrando en fuseWith" << std::endl;

    std::vector<bool> otherCorrs(other.size(), false);
    TMatchingPairList corrs;
    TMatchingPairList::iterator corrIt;
    float corrsRatio;
    CLandmark *thisLM, *otherLM;
    int i, n;
    bool verbose = true;  // false;
    TTimeStamp lastestTime;
    unsigned int nRemoved = 0;

    if (!justInsertAllOfThem)
    {
        // 1) Compute matching between the global and the new map:
        // ---------------------------------------------------------
        computeMatchingWith3DLandmarks(&other, corrs, corrsRatio, otherCorrs);

        // 2) Fuse correspondences
        // ---------------------------------------------------------
        for (corrIt = corrs.begin(); corrIt != corrs.end(); corrIt++)
        {
            thisLM = landmarks.get(corrIt->this_idx);
            otherLM = other.landmarks.get(corrIt->other_idx);

            // Fuse their poses:
            CPointPDFGaussian P, P1, P2;

            thisLM->getPose(P1);
            otherLM->getPose(P2);

            P.bayesianFusion(P1, P2);

            landmarks.isToBeModified(corrIt->this_idx);
            thisLM->setPose(P);

            // Update "seen" data:
            thisLM->seenTimesCount++;
            thisLM->timestampLastSeen = otherLM->timestampLastSeen;

            landmarks.hasBeenModified(corrIt->this_idx);

        }  // end foreach corrs
    }

    // 3) Add new landmarks from the other map:
    // ---------------------------------------------------------
    n = other.size();
    for (i = 0; i < n; i++)
    {
        // Find the lastest time.
        lastestTime =
            max(lastestTime, other.landmarks.get(i)->timestampLastSeen);

        if (!otherCorrs[i])
        {
            // No corrs: A NEW LANDMARK!
            landmarks.push_back(*other.landmarks.get(i));
        }
    }  // end foreach other landmark

    if (!justInsertAllOfThem)
    {
        // 4) Remove landmarks that have been not seen the required
        //      number of times:
        // ---------------------------------------------------------
        n = landmarks.size();
        for (i = n - 1; i >= 0; i--)
        {
            if (landmarks.get(i)->getType() !=
                featNotDefined)  // Occupancy features
            {
                const double dt =
                    1e-3 *
                    std::chrono::duration_cast<std::chrono::milliseconds>(
                        lastestTime - landmarks.get(i)->timestampLastSeen)
                        .count();
                if (dt > fuseOptions.ellapsedTime &&
                    landmarks.get(i)->seenTimesCount < fuseOptions.minTimesSeen)
                {
                    landmarks.erase(i);
                    nRemoved++;
                }
            }
        }
    }

    if (verbose)
    {
        printf(
            "[CLandmarksMap::fuseWith] %u fused/ %u new/ %u removed -> %u "
            "total\n",
            static_cast<unsigned int>(corrs.size()),
            static_cast<unsigned int>(other.size() - corrs.size()),
            static_cast<unsigned int>(nRemoved),
            static_cast<unsigned int>(landmarks.size()));
        FILE* f = os::fopen("Fused.txt", "at");
        fprintf(
            f, "%u\t%u\t%u\t%u\n", static_cast<unsigned int>(corrs.size()),
            static_cast<unsigned int>(other.size() - corrs.size()),
            static_cast<unsigned int>(nRemoved),
            static_cast<unsigned int>(landmarks.size()));
        os::fclose(f);
    }

    MRPT_END
}

/*---------------------------------------------------------------
                        computeMatchingWith3DLandmarks
  ---------------------------------------------------------------*/
void CLandmarksMap::computeMatchingWith3DLandmarks(
    const mrpt::maps::CLandmarksMap* anotherMap,
    TMatchingPairList& correspondences, float& correspondencesRatio,
    std::vector<bool>& otherCorrespondences) const
{
    MRPT_START

    TSequenceLandmarks::const_iterator thisIt, otherIt;
    unsigned int nThis, nOther;
    int maxIdx;
    float desc;
    unsigned int i, n, j, k;
    TMatchingPair match;
    double lik_dist, lik_desc, lik, maxLik;
    // double                                   maxLikDist = -1, maxLikDesc =
    // -1;
    CPointPDFGaussian pointPDF_k, pointPDF_j;
    std::vector<bool> thisLandmarkAssigned;
    double K_desc = 0.0;
    double K_dist = 0.0;

    //  FILE                                    *f = os::fopen( "flik.txt", "wt"
    //);

    // Get the number of landmarks:
    nThis = landmarks.size();
    nOther = anotherMap->landmarks.size();

    // Initially no LM has a correspondence:
    thisLandmarkAssigned.resize(nThis, false);

    // Initially, set all landmarks without correspondences:
    correspondences.clear();
    otherCorrespondences.clear();
    otherCorrespondences.resize(nOther, false);
    correspondencesRatio = 0;

    // Method selection:
    // 0. Our Method.
    // 1. Sim, Elinas, Griffin, Little.

    switch (insertionOptions.SIFTMatching3DMethod)
    {
        case 0:
            // Our method: Filter out by the likelihood of the 3D position and
            // compute the likelihood of the Euclidean descriptor distance

            // "Constants" for the distance computation
            K_desc =
                -0.5 / square(likelihoodOptions.SIFTs_sigma_descriptor_dist);
            K_dist = -0.5 / square(likelihoodOptions.SIFTs_mahaDist_std);

            // CDynamicGrid<std::vector<int32_t>>       *gridLandmarks =
            // landmarks.getGrid();
            // std::vector<int32_t>                     closeLandmarksList;

            for (k = 0, otherIt = anotherMap->landmarks.begin();
                 otherIt != anotherMap->landmarks.end(); otherIt++, k++)
            {
                // Load into "pointPDF_k" the PDF of landmark "otherIt":
                otherIt->getPose(pointPDF_k);

                if (otherIt->getType() == featSIFT)
                {
                    // minDist = minDist2 = 1e10f;
                    maxLik = -1;
                    maxIdx = -1;

                    for (j = 0, thisIt = landmarks.begin();
                         thisIt != landmarks.end(); thisIt++, j++)
                    {
                        if (thisIt->getType() == featSIFT &&
                            thisIt->features.size() ==
                                otherIt->features.size() &&
                            !thisIt->features.empty() &&
                            thisIt->features[0].descriptors.SIFT->size() ==
                                otherIt->features[0].descriptors.SIFT->size())
                        {
                            // Compute "coincidence probability":
                            // --------------------------------------
                            // Load into "pointPDF_j" the PDF of landmark
                            // "otherIt":
                            thisIt->getPose(pointPDF_j);

                            // Compute lik:
                            // lik_dist =
                            // pointPDF_k.productIntegralNormalizedWith(
                            // &pointPDF_j );
                            CMatrixDouble dij(1, 3), Cij(3, 3), Cij_1;
                            double distMahaFlik2;

                            // Distance between means:
                            dij(0, 0) =
                                pointPDF_k.mean.x() - pointPDF_j.mean.x();
                            dij(0, 1) =
                                pointPDF_k.mean.y() - pointPDF_j.mean.y();
                            dij(0, 2) =
                                pointPDF_k.mean.z() - pointPDF_j.mean.z();

                            // Equivalent covariance from "i" to "j":
                            Cij =
                                CMatrixDouble(pointPDF_k.cov + pointPDF_j.cov);
                            Cij_1 = Cij.inverse_LLt();

                            distMahaFlik2 =
                                mrpt::math::multiply_HCHt_scalar(dij, Cij_1);

                            lik_dist = exp(K_dist * distMahaFlik2);
                            // Likelihood regarding the spatial distance

                            if (lik_dist > 1e-2)
                            {
                                // Compute distance between descriptors:
                                // --------------------------------------

                                // MODIFICATION 19-SEPT-2007
                                // ONLY COMPUTE THE EUCLIDEAN DISTANCE BETWEEN
                                // DESCRIPTORS IF IT HAS NOT BEEN COMPUTED
                                // BEFORE
                                // Make the pair of points
                                std::pair<
                                    mrpt::maps::CLandmark::TLandmarkID,
                                    mrpt::maps::CLandmark::TLandmarkID>
                                    mPair(thisIt->ID, otherIt->ID);

                                if (CLandmarksMap::_mEDD[mPair] == 0)
                                {
                                    n = otherIt->features[0]
                                            .descriptors.SIFT->size();
                                    desc = 0;
                                    for (i = 0; i < n; i++)
                                        desc += square(
                                            (*otherIt->features[0]
                                                  .descriptors.SIFT)[i] -
                                            (*thisIt->features[0]
                                                  .descriptors.SIFT)[i]);

                                    CLandmarksMap::_mEDD[mPair] = desc;
                                }  // end if
                                else
                                {
                                    desc = CLandmarksMap::_mEDD[mPair];
                                }

                                lik_desc = exp(K_desc * desc);  // Likelihood
                                // regarding the
                                // descriptor
                            }
                            else
                            {
                                lik_desc = 1e-3;
                            }

                            // Likelihood of the correspondence
                            // --------------------------------------
                            lik = lik_dist * lik_desc;

                            //                      os::fprintf( f,
                            //"%i\t%i\t%f\t%f\t%f\n", k, j, lik_desc, lik_dist,
                            // lik );

                            if (lik > maxLik)
                            {
                                //                          maxLikDist =
                                // lik_dist;
                                //                          maxLikDesc =
                                // lik_desc;
                                maxLik = lik;
                                maxIdx = j;
                            }

                        }  // end of this landmark is SIFT

                    }  // End of for each "this", j
                    // os::fprintf(f, "%i\t%i\t%f\t%f\t%f\n", maxIdx, k,
                    // maxLikDist, maxLikDesc, maxLik);

                    // Is it a correspondence?
                    if (maxLik > insertionOptions.SiftLikelihoodThreshold)
                    {
                        // If a previous correspondence for this LM was found,
                        // discard this one!
                        if (!thisLandmarkAssigned[maxIdx])
                        {
                            thisLandmarkAssigned[maxIdx] = true;

                            // OK: A correspondence found!!
                            otherCorrespondences[k] = true;

                            match.this_idx = maxIdx;
                            match.this_x = landmarks.get(maxIdx)->pose_mean.x;
                            match.this_y = landmarks.get(maxIdx)->pose_mean.y;
                            match.this_z = landmarks.get(maxIdx)->pose_mean.z;

                            match.other_idx = k;
                            match.other_x =
                                anotherMap->landmarks.get(k)->pose_mean.x;
                            match.other_y =
                                anotherMap->landmarks.get(k)->pose_mean.y;
                            match.other_z =
                                anotherMap->landmarks.get(k)->pose_mean.z;

                            correspondences.push_back(match);
                        }
                    }

                }  // end of "otherIt" is SIFT

            }  // end of other it., k

            // Compute the corrs ratio:
            correspondencesRatio =
                correspondences.size() / static_cast<float>(nOther);
            //      os::fclose(f);

            break;

        case 1:

            // IMPLEMENTATION OF THE METHOD DESCRIBED IN [VISION-BASED SLAM
            // USING THE RBPF][SIM, ELINAS, GRIFFIN, LITTLE]
            // 1. Compute Euclidean descriptor distance (EDD).
            // 2. Matching based on a Threshold.
            // 3. Compute likelihood based only on the position of the 3D
            // landmarks.

            // 1.- COMPUTE EDD

            ASSERT_(!anotherMap->landmarks.begin()->features.empty());
            ASSERT_(!landmarks.begin()->features.empty());
            unsigned int dLen = anotherMap->landmarks.begin()
                                    ->features[0]
                                    .descriptors.SIFT->size();
            for (k = 0, otherIt = anotherMap->landmarks.begin();
                 otherIt != anotherMap->landmarks.end(); otherIt++, k++)
            {
                double mEDD = -1.0;
                unsigned int mEDDidx = 0;
                for (j = 0, thisIt = landmarks.begin();
                     thisIt != landmarks.end(); thisIt++, j++)
                {
                    // Compute EDD
                    double EDD = 0.0;
                    for (i = 0; i < dLen; i++)
                        EDD += square(
                            (*otherIt->features[0].descriptors.SIFT)[i] -
                            (*thisIt->features[0].descriptors.SIFT)[i]);

                    EDD = sqrt(EDD);

                    if (EDD > mEDD)
                    {
                        mEDD = EDD;
                        mEDDidx = j;
                    }  // end if
                }  // end for j

                if (mEDD > insertionOptions.SiftEDDThreshold)
                {
                    // There is a correspondence
                    if (!thisLandmarkAssigned[mEDDidx])  // If there is not
                    // multiple
                    // correspondence
                    {
                        thisLandmarkAssigned[mEDDidx] = true;

                        // OK: A correspondence found!!
                        otherCorrespondences[k] = true;

                        otherIt->getPose(pointPDF_k);
                        thisIt->getPose(pointPDF_j);

                        match.this_idx = j;
                        match.this_x = landmarks.get(mEDDidx)->pose_mean.x;
                        match.this_y = landmarks.get(mEDDidx)->pose_mean.y;
                        match.this_z = landmarks.get(mEDDidx)->pose_mean.z;

                        match.other_idx = k;
                        match.other_x =
                            anotherMap->landmarks.get(k)->pose_mean.x;
                        match.other_y =
                            anotherMap->landmarks.get(k)->pose_mean.y;
                        match.other_z =
                            anotherMap->landmarks.get(k)->pose_mean.z;

                        correspondences.push_back(match);

                    }  // end if multiple correspondence

                }  // end if mEDD

            }  // end for k

            correspondencesRatio =
                correspondences.size() / static_cast<float>(nOther);

            break;

    }  // end switch

    MRPT_END
}

/*---------------------------------------------------------------
                        saveToTextFile
  ---------------------------------------------------------------*/
bool CLandmarksMap::saveToTextFile(std::string file)
{
    MRPT_START

    FILE* f = os::fopen(file.c_str(), "wt");
    if (!f) return false;

    // os::fprintf(f,"%% Map of landmarks - file dumped by
    // mrpt::maps::CLandmarksMap\n");
    // os::fprintf(f,"%%  Columns are: X Y Z TYPE(TKeyPointMethod) TIMES_SEEN
    // TIME_OF_LAST_OBSERVATION [SIFT DESCRIPTOR] ID\n");
    // os::fprintf(f,"%%
    // -----------------------------------------------------------------------------------------------------\n");

    for (auto it = landmarks.begin(); it != landmarks.end(); ++it)
    {
        os::fprintf(
            f, "%10f %10f %10f %4i %4u %10f", it->pose_mean.x, it->pose_mean.y,
            it->pose_mean.z, static_cast<int>(it->getType()),
            it->seenTimesCount,
            it->timestampLastSeen == INVALID_TIMESTAMP
                ? 0
                : mrpt::system::extractDayTimeFromTimestamp(
                      it->timestampLastSeen));

        if (it->getType() == featSIFT)
        {
            ASSERT_(!it->features.empty());
            for (unsigned char i : *it->features[0].descriptors.SIFT)
                os::fprintf(f, " %u ", i);
        }
        os::fprintf(f, " %i ", (int)it->ID);

        os::fprintf(f, "\n");
    }

    os::fclose(f);

    return true;

    MRPT_END
}

/*---------------------------------------------------------------
                        saveToMATLABScript3D
  ---------------------------------------------------------------*/
bool CLandmarksMap::saveToMATLABScript3D(
    std::string file, const char* style, float confInterval) const
{
    FILE* f = os::fopen(file.c_str(), "wt");
    if (!f) return false;

    // Header:
    os::fprintf(
        f, "%%-------------------------------------------------------\n");
    os::fprintf(f, "%% File automatically generated using the MRPT method:\n");
    os::fprintf(f, "%%   'CLandmarksMap::saveToMATLABScript3D'\n");
    os::fprintf(f, "%%\n");
    os::fprintf(f, "%%                        ~ MRPT ~\n");
    os::fprintf(
        f, "%%  Jose Luis Blanco Claraco, University of Malaga @ 2006\n");
    os::fprintf(f, "%%  http://www.isa.uma.es/ \n");
    os::fprintf(
        f, "%%-------------------------------------------------------\n\n");

    // Main code:
    os::fprintf(f, "hold on;\n\n");

    for (const auto& landmark : landmarks)
    {
        os::fprintf(
            f, "m=[%.4f %.4f %.4f];", landmark.pose_mean.x,
            landmark.pose_mean.y, landmark.pose_mean.z);
        os::fprintf(
            f, "c=[%.8f %.8f %.8f;%.8f %.8f %.8f;%.8f %.8f %.8f]; ",
            landmark.pose_cov_11, landmark.pose_cov_12, landmark.pose_cov_13,
            landmark.pose_cov_12, landmark.pose_cov_22, landmark.pose_cov_23,
            landmark.pose_cov_13, landmark.pose_cov_23, landmark.pose_cov_33);

        os::fprintf(
            f,
            "error_ellipse(c,m,'conf',%f,'style','%s','numPointsEllipse',10);"
            "\n",
            confInterval, style);
    }

    os::fprintf(f, "axis equal;grid on;xlabel('x'); ylabel('y'); zlabel('z');");

    os::fclose(f);
    return true;
}
/**/

/*---------------------------------------------------------------
                        saveToMATLABScript2D
  ---------------------------------------------------------------*/
bool CLandmarksMap::saveToMATLABScript2D(
    std::string file, const char* style, float stdCount)
{
    FILE* f = os::fopen(file.c_str(), "wt");
    if (!f) return false;

    const int ELLIPSE_POINTS = 30;
    std::vector<float> X, Y, COS, SIN;
    std::vector<float>::iterator x, y, Cos, Sin;
    double ang;
    CMatrixDouble22 cov, eigVal, eigVec, M;

    X.resize(ELLIPSE_POINTS);
    Y.resize(ELLIPSE_POINTS);
    COS.resize(ELLIPSE_POINTS);
    SIN.resize(ELLIPSE_POINTS);

    // Fill the angles:
    for (Cos = COS.begin(), Sin = SIN.begin(), ang = 0; Cos != COS.end();
         Cos++, Sin++, ang += (M_2PI / (ELLIPSE_POINTS - 1)))
    {
        *Cos = cos(ang);
        *Sin = sin(ang);
    }

    // Header:
    os::fprintf(
        f, "%%-------------------------------------------------------\n");
    os::fprintf(f, "%% File automatically generated using the MRPT method:\n");
    os::fprintf(f, "%%   'CLandmarksMap::saveToMATLABScript2D'\n");
    os::fprintf(f, "%%\n");
    os::fprintf(f, "%%                        ~ MRPT ~\n");
    os::fprintf(
        f, "%%  Jose Luis Blanco Claraco, University of Malaga @ 2006\n");
    os::fprintf(f, "%%  http://www.isa.uma.es/ \n");
    os::fprintf(
        f, "%%-------------------------------------------------------\n\n");

    // Main code:
    os::fprintf(f, "hold on;\n\n");

    for (auto& landmark : landmarks)
    {
        // Compute the eigen-vectors & values:
        cov(0, 0) = landmark.pose_cov_11;
        cov(1, 1) = landmark.pose_cov_22;
        cov(0, 1) = cov(1, 0) = landmark.pose_cov_12;

        std::vector<double> eigvals;
        cov.eig_symmetric(eigVec, eigvals);
        eigVal.setZero();
        eigVal.setDiagonal(eigvals);
        eigVal = eigVal.array().sqrt().matrix();
        M = eigVal.asEigen() * eigVec.transpose();

        // Compute the points of the ellipsoid:
        // ----------------------------------------------
        for (x = X.begin(), y = Y.begin(), Cos = COS.begin(), Sin = SIN.begin();
             x != X.end(); x++, y++, Cos++, Sin++)
        {
            *x =
                (landmark.pose_mean.x +
                 stdCount * (*Cos * M(0, 0) + *Sin * M(1, 0)));
            *y =
                (landmark.pose_mean.y +
                 stdCount * (*Cos * M(0, 1) + *Sin * M(1, 1)));
        }

        // Save the code to plot the ellipsoid:
        // ----------------------------------------------
        os::fprintf(f, "plot([ ");
        for (x = X.begin(); x != X.end(); x++)
        {
            os::fprintf(f, "%.4f", *x);
            if (x != (X.end() - 1)) os::fprintf(f, ",");
        }
        os::fprintf(f, "],[ ");
        for (y = Y.begin(); y != Y.end(); y++)
        {
            os::fprintf(f, "%.4f", *y);
            if (y != (Y.end() - 1)) os::fprintf(f, ",");
        }

        os::fprintf(f, "],'%s');\n", style);

        // os::fprintf(f,"error_ellipse(c,m,'conf',0.99,'style','%s','numPointsEllipse',10);\n",style);
    }

    os::fprintf(f, "\naxis equal;\n");
    os::fclose(f);
    return true;
}

/*---------------------------------------------------------------
                        loadOccupancyFeaturesFrom2DRangeScan
  ---------------------------------------------------------------*/
void CLandmarksMap::loadOccupancyFeaturesFrom2DRangeScan(
    const CObservation2DRangeScan& obs, const CPose3D* robotPose,
    unsigned int downSampleFactor)
{
    unsigned int i, n = obs.scan.size();
    double Th, dTh;  // angle of the ray
    double J11, J12, J21, J22;  // The jacobian elements.
    double d;
    CPose3D sensorGlobalPose;

    // Empty the map:
    this->clear();

    // Sensor pose in 3D:
    if (!robotPose)
        sensorGlobalPose = obs.sensorPose;
    else
        sensorGlobalPose = *robotPose + obs.sensorPose;

    // Starting direction:
    if (obs.rightToLeft)
    {
        Th = -0.5 * obs.aperture;
        dTh = obs.aperture / n;
    }
    else
    {
        Th = +0.5 * obs.aperture;
        dTh = -obs.aperture / n;
    }

    // Measurement uncertainties:
    double var_d = square(0.005);  // square(obs.stdError);
    double var_th = square(dTh / 10.0);

    // For each range:
    for (i = 0; i < n; i += downSampleFactor, Th += downSampleFactor * dTh)
    {
        // If it is a valid ray:
        if (obs.validRange[i])
        {
            CLandmark newLandmark;

            // Timestamp:
            newLandmark.timestampLastSeen = obs.timestamp;
            newLandmark.seenTimesCount = 1;

            newLandmark.createOneFeature();
            newLandmark.features[0].type = featNotDefined;

            d = obs.scan[i];

            // Compute the landmark in 2D:
            // -----------------------------------------------
            // Descriptor:
            newLandmark.features[0].orientation = Th;
            newLandmark.features[0].keypoint.octave = d;

            // Mean:
            newLandmark.pose_mean.x = (cos(Th) * d);
            newLandmark.pose_mean.y = (sin(Th) * d);
            newLandmark.pose_mean.z = 0;

            // Normal direction:
            newLandmark.normal = newLandmark.pose_mean;
            newLandmark.normal *= -1.0f / newLandmark.pose_mean.norm();

            // Jacobian:
            J11 = -d * sin(Th);
            J12 = cos(Th);
            J21 = d * cos(Th);
            J22 = sin(Th);

            // Covariance matrix:
            newLandmark.pose_cov_11 = (J11 * J11 * var_th + J12 * J12 * var_d);
            newLandmark.pose_cov_12 = (J11 * J21 * var_th + J12 * J22 * var_d);
            newLandmark.pose_cov_22 = (J21 * J21 * var_th + J22 * J22 * var_d);
            newLandmark.pose_cov_33 = (square(0.005));  // var_d;
            newLandmark.pose_cov_13 = newLandmark.pose_cov_23 = 0;

            // Append it:
            // -----------------------------------------------
            landmarks.push_back(newLandmark);

        }  // end of valid ray.

    }  // end for n

    // Take landmarks to 3D according to the robot & sensor 3D poses:
    // -----------------------------------------------
    changeCoordinatesReference(sensorGlobalPose);
}

/*---------------------------------------------------------------
                METHOD DESCRIBED IN PAPER
                -------------------------

        ICRA 2007,....

  ---------------------------------------------------------------*/
double CLandmarksMap::computeLikelihood_RSLC_2007(
    const CLandmarksMap* s, const CPose2D& sensorPose)
{
    MRPT_UNUSED_PARAM(sensorPose);
    MRPT_START

    double lik = 1.0;
    TSequenceLandmarks::const_iterator itOther;
    CLandmark* lm;  //*itClosest;
    double corr;
    double PrNoCorr;
    CPointPDFGaussian poseThis, poseOther;
    // double                               STD_THETA = DEG2RAD(0.15);
    // double                               STD_DIST = 0.5f;
    double nFoundCorrs = 0;
    std::vector<int32_t>* corrs;
    unsigned int cx, cy, cx_1, cx_2, cy_1, cy_2;

    //  s->saveToMATLABScript2D(std::string("ver_sensed.m"));
    // saveToMATLABScript2D(std::string("ver_ref.m"),"r");

    CDynamicGrid<std::vector<int32_t>>* grid = landmarks.getGrid();
    // grid->saveToTextFile( "debug_grid.txt" );

    // For each landmark in the observations:
    for (itOther = s->landmarks.begin(); itOther != s->landmarks.end();
         itOther++)
    {
        itOther->getPose(poseOther);

        cx = cy = grid->y2idx(itOther->pose_mean.y);

        cx_1 = max(0, grid->x2idx(itOther->pose_mean.x - 0.10f));
        cx_2 =
            min(static_cast<int>(grid->getSizeX()) - 1,
                grid->x2idx(itOther->pose_mean.x + 0.10f));
        cy_1 = max(0, grid->y2idx(itOther->pose_mean.y - 0.10f));
        cy_2 =
            min(static_cast<int>(grid->getSizeY()) - 1,
                grid->y2idx(itOther->pose_mean.y + 0.10f));

        // The likelihood of no correspondence starts at "1" and will be
        // modified next:
        PrNoCorr = 1;

        // For each landmark in this map: Compute its correspondence likelihood
        //   and conditioned observation likelihood:
        // itClosest = nullptr;

        // Look at landmarks in sourronding cells only:
        for (cx = cx_1; cx <= cx_2; cx++)
            for (cy = cy_1; cy <= cy_2; cy++)
            {
                corrs = grid->cellByIndex(cx, cy);
                ASSERT_(corrs != nullptr);
                if (!corrs->empty())
                    for (int& it : *corrs)
                    {
                        lm = landmarks.get(it);

                        // Compute the "correspondence" in the range [0,1]:
                        // -------------------------------------------------------------

                        // Shortcut:
                        if (fabs(lm->pose_mean.x - itOther->pose_mean.x) >
                                0.15f ||
                            fabs(lm->pose_mean.y - itOther->pose_mean.y) >
                                0.15f)
                        {
                            // Absolutely no correspondence, not need to compute
                            // it:
                            corr = 0;
                        }
                        else
                        {
                            lm->getPose(poseThis);
                            corr = poseOther.productIntegralNormalizedWith2D(
                                poseThis);
                        }

                        // The likelihood of no corresp. is proportional to the
                        // product of all "1-CORRij":
                        // -------------------------------------------------------------
                        PrNoCorr *= 1 - corr;

                    }  // end of each landmark in this map.
            }

        // Only consider this "point" if it has some real chance to has a
        // correspondence:
        nFoundCorrs += 1 - PrNoCorr;

        /**** DEBUG **** /
        {
            //FILE  *f=os::fopen("debug_corrs.txt","wt");
            //for (Cij=v_Cij.begin(),pZj=v_pZj.begin(); pZj!=v_pZj.end();
        Cij++,pZj++)
            //{
            //  os::fprintf(f,"%e %e\n", *Cij, *pZj);
            //}

            //os::fprintf(f,"\n INDIV LIK=%e\n lik=%e\n
        closestObstacleInLine=%e\n measured
        range=%e\n",indivLik,lik,closestObstacleInLine,
        itOther.descriptors.SIFT[1]);
            //os::fprintf(f,"
        closestObstacleDirection=%e\n",closestObstacleDirection);
            //os::fclose(f);

            printf("\n lik=%e\n closestObstacleInLine=%e\n measured
        range=%e\n",lik,closestObstacleInLine, itOther.descriptors.SIFT[1]);
            if (itClosest)
                    printf(" closest=(%.03f,%.03f)\n", itClosest->pose_mean.x,
        itClosest->pose_mean.y);
            else    printf(" closest=nullptr\n");

            printf(" P(no corrs)=%e\n", PrNoCorr );
            mrpt::system::pause();
        }
        / ***************/

        lik *= 1 - PrNoCorr;

    }  // enf for each landmark in the other map.

    lik = nFoundCorrs / static_cast<double>(s->size());

    lik = log(lik);

    MRPT_CHECK_NORMAL_NUMBER(lik);
    return lik;

    MRPT_END
}

/*---------------------------------------------------------------

                    TCustomSequenceLandmarks

  ---------------------------------------------------------------*/
CLandmarksMap::TCustomSequenceLandmarks::TCustomSequenceLandmarks()
    : m_landmarks(), m_grid(-10.0f, 10.0f, -10.0f, 10.f, 0.20f)

{
}

void CLandmarksMap::TCustomSequenceLandmarks::clear()
{
    m_landmarks.clear();

    // Erase the grid:
    m_grid.clear();

    m_largestDistanceFromOriginIsUpdated = false;
}

void CLandmarksMap::TCustomSequenceLandmarks::push_back(const CLandmark& l)
{
    // Resize grid if necesary:
    std::vector<int32_t> dummyEmpty;

    m_grid.resize(
        min(m_grid.getXMin(), l.pose_mean.x - 0.1),
        max(m_grid.getXMax(), l.pose_mean.x + 0.1),
        min(m_grid.getYMin(), l.pose_mean.y - 0.1),
        max(m_grid.getYMax(), l.pose_mean.y + 0.1), dummyEmpty);

    m_landmarks.push_back(l);

    // Add to the grid:
    std::vector<int32_t>* cell = m_grid.cellByPos(l.pose_mean.x, l.pose_mean.y);
    ASSERT_(cell);
    cell->push_back(m_landmarks.size() - 1);

    m_largestDistanceFromOriginIsUpdated = false;
}

CLandmark* CLandmarksMap::TCustomSequenceLandmarks::get(unsigned int indx)
{
    return &m_landmarks[indx];
}

const CLandmark* CLandmarksMap::TCustomSequenceLandmarks::get(
    unsigned int indx) const
{
    return &m_landmarks[indx];
}

void CLandmarksMap::TCustomSequenceLandmarks::isToBeModified(unsigned int indx)
{
    std::vector<int32_t>* cell = m_grid.cellByPos(
        m_landmarks[indx].pose_mean.x, m_landmarks[indx].pose_mean.y);

    std::vector<int32_t>::iterator it;
    for (it = cell->begin(); it != cell->end(); it++)
    {
        if (*it == static_cast<int>(indx))
        {
            cell->erase(it);
            return;
        }
    }

    m_largestDistanceFromOriginIsUpdated = false;
}

void CLandmarksMap::TCustomSequenceLandmarks::erase(unsigned int indx)
{
    m_landmarks.erase(m_landmarks.begin() + indx);
    m_largestDistanceFromOriginIsUpdated = false;
}

void CLandmarksMap::TCustomSequenceLandmarks::hasBeenModified(unsigned int indx)
{
    std::vector<int32_t> dummyEmpty;

    // Resize grid if necesary:
    m_grid.resize(
        min(m_grid.getXMin(), m_landmarks[indx].pose_mean.x),
        max(m_grid.getXMax(), m_landmarks[indx].pose_mean.x),
        min(m_grid.getYMin(), m_landmarks[indx].pose_mean.y),
        max(m_grid.getYMax(), m_landmarks[indx].pose_mean.y), dummyEmpty);

    // Add to the grid:
    std::vector<int32_t>* cell = m_grid.cellByPos(
        m_landmarks[indx].pose_mean.x, m_landmarks[indx].pose_mean.y);
    cell->push_back(indx);
    m_largestDistanceFromOriginIsUpdated = false;
}

void CLandmarksMap::TCustomSequenceLandmarks::hasBeenModifiedAll()
{
    MRPT_START

    TSequenceLandmarks::iterator it;
    unsigned int idx;
    double min_x = -10.0, max_x = 10.0;
    double min_y = -10.0, max_y = 10.0;
    std::vector<int32_t> dummyEmpty;

    // Clear cells:
    m_grid.clear();

    // Resize the grid to the outer limits of landmarks:
    for (idx = 0, it = m_landmarks.begin(); it != m_landmarks.end();
         idx++, it++)
    {
        min_x = min(min_x, it->pose_mean.x);
        max_x = max(max_x, it->pose_mean.x);
        min_y = min(min_y, it->pose_mean.y);
        max_y = max(max_y, it->pose_mean.y);
    }
    m_grid.resize(min_x, max_x, min_y, max_y, dummyEmpty);

    // Add landmarks to cells:
    for (idx = 0, it = m_landmarks.begin(); it != m_landmarks.end();
         idx++, it++)
    {
        std::vector<int32_t>* cell =
            m_grid.cellByPos(it->pose_mean.x, it->pose_mean.y);
        cell->push_back(idx);
    }

    m_largestDistanceFromOriginIsUpdated = false;
    MRPT_END
}

/*---------------------------------------------------------------
                        getLargestDistanceFromOrigin
---------------------------------------------------------------*/
float CLandmarksMap::TCustomSequenceLandmarks::getLargestDistanceFromOrigin()
    const
{
    // Updated?
    if (!m_largestDistanceFromOriginIsUpdated)
    {
        // NO: Update it:
        float maxDistSq = 0, d;
        for (const auto& it : *this)
        {
            d = square(it.pose_mean.x) + square(it.pose_mean.y) +
                square(it.pose_mean.z);
            maxDistSq = max(d, maxDistSq);
        }

        m_largestDistanceFromOrigin = sqrt(maxDistSq);
        m_largestDistanceFromOriginIsUpdated = true;
    }
    return m_largestDistanceFromOrigin;
}

/*---------------------------------------------------------------
                    computeLikelihood_SIFT_LandmarkMap
  ---------------------------------------------------------------*/
double CLandmarksMap::computeLikelihood_SIFT_LandmarkMap(
    CLandmarksMap* theMap, TMatchingPairList* correspondences,
    std::vector<bool>* otherCorrespondences)
{
    double lik = 0;  // For 'traditional'
    double lik_i;
    unsigned long distDesc;
    double likByDist, likByDesc;

    std::vector<unsigned char>::iterator it1, it2;
    double K_dist = -0.5 / square(likelihoodOptions.SIFTs_mahaDist_std);
    double K_desc =
        -0.5 / square(likelihoodOptions.SIFTs_sigma_descriptor_dist);

    unsigned int idx1, idx2;
    CPointPDFGaussian lm1_pose, lm2_pose;
    CMatrixD dij(1, 3), Cij(3, 3), Cij_1;
    double distMahaFlik2;
    int decimation = likelihoodOptions.SIFTs_decimation;

    // USE INSERTOPTIONS METHOD
    switch (insertionOptions.SIFTLikelihoodMethod)
    {
        case 0:  // Our method
        {
            // lik = 1e-9;      // For consensus
            lik = 1.0;  // For traditional

            TSequenceLandmarks::iterator lm1, lm2;
            for (idx1 = 0, lm1 = theMap->landmarks.begin();
                 lm1 < theMap->landmarks.end();
                 lm1 += decimation, idx1 += decimation)  // Other theMap LM1
            {
                if (lm1->getType() == featSIFT)
                {
                    // Get the pose of lm1 as an object:
                    lm1->getPose(lm1_pose);

                    lik_i = 0;  // Counter

                    for (idx2 = 0, lm2 = landmarks.begin();
                         lm2 != landmarks.end();
                         lm2++, idx2++)  // This theMap LM2
                    {
                        if (lm2->getType() == featSIFT)
                        {
                            // Get the pose of lm2 as an object:
                            lm2->getPose(lm2_pose);

                            // Compute the likelihood according to mahalanobis
                            // distance:
                            // Distance between means:
                            dij(0, 0) = lm1->pose_mean.x - lm2->pose_mean.x;
                            dij(0, 1) = lm1->pose_mean.y - lm2->pose_mean.y;
                            dij(0, 2) = lm1->pose_mean.z - lm2->pose_mean.z;

                            // std::cout << "ED POSICION: " << sqrt(
                            // dij(0,0)*dij(0,0) + dij(0,1)*dij(0,1) +
                            // dij(0,2)*dij(0,2) ) << std::endl;
                            // Equivalent covariance from "i" to "j":
                            Cij = CMatrixDouble(lm1_pose.cov + lm2_pose.cov);
                            Cij_1 = Cij.inverse_LLt();

                            distMahaFlik2 =
                                mrpt::math::multiply_HCHt_scalar(dij, Cij_1);

                            likByDist = exp(K_dist * distMahaFlik2);

                            if (likByDist > 1e-2)
                            {
                                // If the EUCLIDEAN distance is not too large,
                                // we compute the Descriptor distance
                                // Compute Likelihood by descriptor
                                // Compute distance between descriptors

                                // IF the EDD has been already computed, we skip
                                // this step!
                                std::pair<
                                    mrpt::maps::CLandmark::TLandmarkID,
                                    mrpt::maps::CLandmark::TLandmarkID>
                                    mPair(lm2->ID, lm1->ID);
                                // std::cout << "Par: (" << lm2->ID << "," <<
                                // lm1->ID << ") -> ";

                                if (CLandmarksMap::_mEDD[mPair] == 0)
                                {
                                    distDesc = 0;
                                    ASSERT_(
                                        !lm1->features.empty() &&
                                        !lm2->features.empty());
                                    ASSERT_(
                                        lm1->features[0]
                                            .descriptors.SIFT->size() ==
                                        lm2->features[0]
                                            .descriptors.SIFT->size());

                                    for (it1 = lm1->features[0]
                                                   .descriptors.SIFT->begin(),
                                        it2 = lm2->features[0]
                                                  .descriptors.SIFT->begin();
                                         it1 != lm1->features[0]
                                                    .descriptors.SIFT->end();
                                         it1++, it2++)
                                        distDesc += square(*it1 - *it2);

                                    // Insert into the vector of Euclidean
                                    // distances
                                    CLandmarksMap::_mEDD[mPair] = distDesc;
                                }  // end if
                                else
                                {
                                    distDesc = (unsigned long)
                                        CLandmarksMap::_mEDD[mPair];
                                }
                                likByDesc = exp(K_desc * distDesc);
                                lik_i += likByDist *
                                         likByDesc;  // Cumulative Likelihood
                            }
                            else
                            {
                                // If the EUCLIDEAN distance is too large, we
                                // assume that the cumulative likelihood is
                                // (almost) zero.
                                lik_i += 1e-10f;
                            }
                        }  // end if
                    }  // end for "lm2"
                    lik *= (0.1 + 0.9 * lik_i);  // (TRADITIONAL) Total
                }
            }  // end for "lm1"
        }
        break;

        case 1:  // SIM, ELINAS, GRIFFIN, LITTLE
            double dist;
            unsigned int k;
            TMatchingPairList::iterator itCorr;

            lik = 1.0f;

            // Check if the Matches are inserted into the function
            if (correspondences == nullptr)
                THROW_EXCEPTION(
                    "When using this method with SIFTLikelihoodMethod = 1, "
                    "TMatchingPairList *correspondence can not be NULL");

            if (otherCorrespondences == nullptr)
                THROW_EXCEPTION(
                    "When using this method with SIFTLikelihoodMethod = 1, "
                    "std::vector<bool> *otherCorrespondences can not be NULL");

            for (itCorr = correspondences->begin();
                 itCorr != correspondences->end(); itCorr++)
            {
                CLandmark* lm1 = theMap->landmarks.get(itCorr->other_idx);
                CLandmark* lm2 = landmarks.get(itCorr->this_idx);

                dij(0, 0) = lm1->pose_mean.x - lm2->pose_mean.x;
                dij(0, 1) = lm1->pose_mean.y - lm2->pose_mean.y;
                dij(0, 2) = lm1->pose_mean.z - lm2->pose_mean.z;

                // Equivalent covariance from "i" to "j":
                Cij = CMatrixDouble(lm1_pose.cov + lm2_pose.cov);
                Cij_1 = Cij.inverse_LLt();

                distMahaFlik2 = mrpt::math::multiply_HCHt_scalar(dij, Cij_1);

                dist = min(
                    (double)likelihoodOptions.SIFTnullCorrespondenceDistance,
                    distMahaFlik2);

                lik *= exp(-0.5 * dist);

            }  // end for correspondences

            // We complete the likelihood with the null correspondences
            for (k = 1; k <= (theMap->size() - correspondences->size()); k++)
                lik *= likelihoodOptions.SIFTnullCorrespondenceDistance;

            break;

    }  // end switch

    lik = log(lik);
    MRPT_CHECK_NORMAL_NUMBER(lik);
    return lik;
}

/*---------------------------------------------------------------
                    TInsertionOptions
  ---------------------------------------------------------------*/
CLandmarksMap::TInsertionOptions::TInsertionOptions()
    :

      SIFT_feat_options(vision::featSIFT)
{
}

/*---------------------------------------------------------------
                    dumpToTextStream
  ---------------------------------------------------------------*/
void CLandmarksMap::TInsertionOptions::dumpToTextStream(std::ostream& out) const
{
    out << mrpt::format(
        "\n----------- [CLandmarksMap::TInsertionOptions] ------------ \n\n");

    out << mrpt::format(
        "insert_SIFTs_from_monocular_images      = %c\n",
        insert_SIFTs_from_monocular_images ? 'Y' : 'N');
    out << mrpt::format(
        "insert_SIFTs_from_stereo_images         = %c\n",
        insert_SIFTs_from_stereo_images ? 'Y' : 'N');
    out << mrpt::format(
        "insert_Landmarks_from_range_scans       = %c\n",
        insert_Landmarks_from_range_scans ? 'Y' : 'N');
    out << mrpt::format("\n");
    out << mrpt::format(
        "SiftCorrRatioThreshold                  = %f\n",
        SiftCorrRatioThreshold);
    out << mrpt::format(
        "SiftLikelihoodThreshold                 = %f\n",
        SiftLikelihoodThreshold);
    out << mrpt::format(
        "SiftEDDThreshold                        = %f\n", SiftEDDThreshold);
    out << mrpt::format(
        "SIFTMatching3DMethod                    = %d\n", SIFTMatching3DMethod);
    out << mrpt::format(
        "SIFTLikelihoodMethod                    = %d\n", SIFTLikelihoodMethod);

    out << mrpt::format(
        "SIFTsLoadDistanceOfTheMean              = %f\n",
        SIFTsLoadDistanceOfTheMean);
    out << mrpt::format(
        "SIFTsLoadEllipsoidWidth                 = %f\n",
        SIFTsLoadEllipsoidWidth);
    out << mrpt::format("\n");
    out << mrpt::format(
        "SIFTs_stdXY                             = %f\n", SIFTs_stdXY);
    out << mrpt::format(
        "SIFTs_stdDisparity                      = %f\n", SIFTs_stdDisparity);
    out << mrpt::format("\n");
    out << mrpt::format(
        "SIFTs_numberOfKLTKeypoints              = %i\n",
        SIFTs_numberOfKLTKeypoints);
    out << mrpt::format(
        "SIFTs_stereo_maxDepth                   = %f\n",
        SIFTs_stereo_maxDepth);
    out << mrpt::format(
        "SIFTs_epipolar_TH                       = %f\n", SIFTs_epipolar_TH);
    out << mrpt::format(
        "PLOT_IMAGES                             = %c\n",
        PLOT_IMAGES ? 'Y' : 'N');

    SIFT_feat_options.dumpToTextStream(out);

    out << mrpt::format("\n");
}

/*---------------------------------------------------------------
                    loadFromConfigFile
  ---------------------------------------------------------------*/
void CLandmarksMap::TInsertionOptions::loadFromConfigFile(
    const mrpt::config::CConfigFileBase& iniFile, const std::string& section)
{
    insert_SIFTs_from_monocular_images = iniFile.read_bool(
        section.c_str(), "insert_SIFTs_from_monocular_images",
        insert_SIFTs_from_monocular_images);
    insert_SIFTs_from_stereo_images = iniFile.read_bool(
        section.c_str(), "insert_SIFTs_from_stereo_images",
        insert_SIFTs_from_stereo_images);
    insert_Landmarks_from_range_scans = iniFile.read_bool(
        section.c_str(), "insert_Landmarks_from_range_scans",
        insert_Landmarks_from_range_scans);
    SiftCorrRatioThreshold = iniFile.read_float(
        section.c_str(), "SiftCorrRatioThreshold", SiftCorrRatioThreshold);
    SiftLikelihoodThreshold = iniFile.read_float(
        section.c_str(), "SiftLikelihoodThreshold", SiftLikelihoodThreshold);
    SiftEDDThreshold = iniFile.read_float(
        section.c_str(), "SiftEDDThreshold", SiftEDDThreshold);
    SIFTMatching3DMethod = iniFile.read_int(
        section.c_str(), "SIFTMatching3DMethod", SIFTMatching3DMethod);
    SIFTLikelihoodMethod = iniFile.read_int(
        section.c_str(), "SIFTLikelihoodMethod", SIFTLikelihoodMethod);
    SIFTsLoadDistanceOfTheMean = iniFile.read_float(
        section.c_str(), "SIFTsLoadDistanceOfTheMean",
        SIFTsLoadDistanceOfTheMean);
    SIFTsLoadEllipsoidWidth = iniFile.read_float(
        section.c_str(), "SIFTsLoadEllipsoidWidth", SIFTsLoadEllipsoidWidth);
    SIFTs_stdXY =
        iniFile.read_float(section.c_str(), "SIFTs_stdXY", SIFTs_stdXY);
    SIFTs_stdDisparity = iniFile.read_float(
        section.c_str(), "SIFTs_stdDisparity", SIFTs_stdDisparity);
    SIFTs_numberOfKLTKeypoints = iniFile.read_int(
        section.c_str(), "SIFTs_numberOfKLTKeypoints",
        SIFTs_numberOfKLTKeypoints);
    SIFTs_stereo_maxDepth = iniFile.read_float(
        section.c_str(), "SIFTs_stereo_maxDepth", SIFTs_stereo_maxDepth);
    SIFTs_epipolar_TH = iniFile.read_float(
        section.c_str(), "SIFTs_epipolar_TH", SIFTs_epipolar_TH);
    PLOT_IMAGES =
        iniFile.read_bool(section.c_str(), "PLOT_IMAGES", PLOT_IMAGES);

    SIFT_feat_options.loadFromConfigFile(iniFile, section);
}

/*---------------------------------------------------------------
                    TLikelihoodOptions
  ---------------------------------------------------------------*/
CLandmarksMap::TLikelihoodOptions::TLikelihoodOptions()
    : SIFT_feat_options(vision::featSIFT),

      GPSOrigin()

{
}

CLandmarksMap::TLikelihoodOptions::TGPSOrigin::TGPSOrigin()

    = default;

/*---------------------------------------------------------------
                    dumpToTextStream
  ---------------------------------------------------------------*/
void CLandmarksMap::TLikelihoodOptions::dumpToTextStream(
    std::ostream& out) const
{
    out << mrpt::format(
        "\n----------- [CLandmarksMap::TLikelihoodOptions] ------------ \n\n");

    out << mrpt::format(
        "rangeScan2D_decimation                  = %i\n",
        rangeScan2D_decimation);
    out << mrpt::format(
        "SIFTs_sigma_euclidean_dist              = %f\n",
        SIFTs_sigma_euclidean_dist);
    out << mrpt::format(
        "SIFTs_sigma_descriptor_dist             = %f\n",
        SIFTs_sigma_descriptor_dist);
    out << mrpt::format(
        "SIFTs_mahaDist_std                      = %f\n", SIFTs_mahaDist_std);
    out << mrpt::format(
        "SIFTs_decimation                        = %i\n", SIFTs_decimation);
    out << mrpt::format(
        "SIFTnullCorrespondenceDistance          = %f\n",
        SIFTnullCorrespondenceDistance);
    out << mrpt::format(
        "beaconRangesStd                         = %f\n", beaconRangesStd);
    out << mrpt::format(
        "beaconRangesUseObservationStd           = %c\n",
        beaconRangesUseObservationStd ? 'Y' : 'N');
    out << mrpt::format(
        "extRobotPoseStd                         = %f\n", extRobotPoseStd);

    out << mrpt::format(
        "GPSOrigin:LATITUDE                      = %f\n", GPSOrigin.latitude);
    out << mrpt::format(
        "GPSOrigin:LONGITUDE                     = %f\n", GPSOrigin.longitude);
    out << mrpt::format(
        "GPSOrigin:ALTITUDE                      = %f\n", GPSOrigin.altitude);
    out << mrpt::format(
        "GPSOrigin:Rotation_Angle                = %f\n", GPSOrigin.ang);
    out << mrpt::format(
        "GPSOrigin:x_shift                       = %f\n", GPSOrigin.x_shift);
    out << mrpt::format(
        "GPSOrigin:y_shift                       = %f\n", GPSOrigin.y_shift);
    out << mrpt::format(
        "GPSOrigin:min_sat                       = %i\n", GPSOrigin.min_sat);

    out << mrpt::format(
        "GPS_sigma                               = %f (m)\n", GPS_sigma);

    SIFT_feat_options.dumpToTextStream(out);

    out << mrpt::format("\n");
}

/*---------------------------------------------------------------
                    loadFromConfigFile
  ---------------------------------------------------------------*/
void CLandmarksMap::TLikelihoodOptions::loadFromConfigFile(
    const mrpt::config::CConfigFileBase& iniFile, const std::string& section)
{
    rangeScan2D_decimation = iniFile.read_int(
        section.c_str(), "rangeScan2D_decimation", rangeScan2D_decimation);
    SIFTs_sigma_euclidean_dist = iniFile.read_double(
        section.c_str(), "SIFTs_sigma_euclidean_dist",
        SIFTs_sigma_euclidean_dist);
    SIFTs_sigma_descriptor_dist = iniFile.read_double(
        section.c_str(), "SIFTs_sigma_descriptor_dist",
        SIFTs_sigma_descriptor_dist);
    SIFTs_mahaDist_std = iniFile.read_float(
        section.c_str(), "SIFTs_mahaDist_std", SIFTs_mahaDist_std);
    SIFTs_decimation =
        iniFile.read_int(section.c_str(), "SIFTs_decimation", SIFTs_decimation);
    SIFTnullCorrespondenceDistance = iniFile.read_float(
        section.c_str(), "SIFTnullCorrespondenceDistance",
        SIFTnullCorrespondenceDistance);

    GPSOrigin.latitude = iniFile.read_double(
        section.c_str(), "GPSOriginLatitude", GPSOrigin.latitude);
    GPSOrigin.longitude = iniFile.read_double(
        section.c_str(), "GPSOriginLongitude", GPSOrigin.longitude);
    GPSOrigin.altitude = iniFile.read_double(
        section.c_str(), "GPSOriginAltitude", GPSOrigin.altitude);
    GPSOrigin.ang =
        iniFile.read_double(section.c_str(), "GPSOriginAngle", GPSOrigin.ang) *
        M_PI / 180;
    GPSOrigin.x_shift = iniFile.read_double(
        section.c_str(), "GPSOriginXshift", GPSOrigin.x_shift);
    GPSOrigin.y_shift = iniFile.read_double(
        section.c_str(), "GPSOriginYshift", GPSOrigin.y_shift);
    GPSOrigin.min_sat =
        iniFile.read_int(section.c_str(), "GPSOriginMinSat", GPSOrigin.min_sat);

    GPS_sigma = iniFile.read_float(section.c_str(), "GPSSigma", GPS_sigma);

    beaconRangesStd =
        iniFile.read_float(section.c_str(), "beaconRangesStd", beaconRangesStd);
    beaconRangesUseObservationStd = iniFile.read_bool(
        section.c_str(), "beaconRangesUseObservationStd",
        beaconRangesUseObservationStd);

    extRobotPoseStd =
        iniFile.read_float(section.c_str(), "extRobotPoseStd", extRobotPoseStd);

    SIFT_feat_options.loadFromConfigFile(iniFile, section);
}

/*---------------------------------------------------------------
                    TFuseOptions
  ---------------------------------------------------------------*/
CLandmarksMap::TFuseOptions::TFuseOptions()

    = default;

/*---------------------------------------------------------------
                     isEmpty
  ---------------------------------------------------------------*/
bool CLandmarksMap::isEmpty() const { return size() == 0; }
/*---------------------------------------------------------------
                     simulateBeaconReadings
  ---------------------------------------------------------------*/
void CLandmarksMap::simulateBeaconReadings(
    const CPose3D& in_robotPose, const CPoint3D& in_sensorLocationOnRobot,
    mrpt::obs::CObservationBeaconRanges& out_Observations) const
{
    TSequenceLandmarks::const_iterator it;
    mrpt::obs::CObservationBeaconRanges::TMeasurement newMeas;
    CPoint3D point3D, beacon3D;
    CPointPDFGaussian beaconPDF;

    // Compute the 3D position of the sensor:
    point3D = in_robotPose + in_sensorLocationOnRobot;

    // Clear output data:
    out_Observations.sensedData.clear();
    out_Observations.timestamp = mrpt::system::getCurrentTime();

    // For each BEACON landmark in the map:
    for (it = landmarks.begin(); it != landmarks.end(); it++)
    {
        if (it->getType() == featBeacon)
        {
            // Get the 3D position of the beacon (just the mean value):
            it->getPose(beaconPDF);
            beacon3D = beaconPDF.mean;

            float range = point3D.distanceTo(beacon3D);

            // Add noise:
            range += out_Observations.stdError *
                     getRandomGenerator().drawGaussian1D_normalized();
            range = max(0.0f, range);

            if (range >= out_Observations.minSensorDistance &&
                range <= out_Observations.maxSensorDistance)
            {
                // Fill out:
                newMeas.beaconID = it->ID;
                newMeas.sensorLocationOnRobot = in_sensorLocationOnRobot;
                newMeas.sensedDistance = range;

                // Insert:
                out_Observations.sensedData.push_back(newMeas);
            }
        }  // end if beacon
    }  // end for it
    // Done!
}

/*---------------------------------------------------------------
                     saveMetricMapRepresentationToFile
  ---------------------------------------------------------------*/
void CLandmarksMap::saveMetricMapRepresentationToFile(
    const std::string& filNamePrefix) const
{
    MRPT_START

    // Matlab:
    std::string fil1(filNamePrefix + std::string("_3D.m"));
    saveToMATLABScript3D(fil1);

    // 3D Scene:
    opengl::COpenGLScene scene;
    mrpt::opengl::CSetOfObjects::Ptr obj3D =
        mrpt::opengl::CSetOfObjects::Create();
    getAs3DObject(obj3D);

    opengl::CGridPlaneXY::Ptr objGround =
        std::make_shared<opengl::CGridPlaneXY>(-100, 100, -100, 100, 0, 1);

    scene.insert(obj3D);
    scene.insert(objGround);

    std::string fil2(filNamePrefix + std::string("_3D.3Dscene"));
    scene.saveToFile(fil2);

    MRPT_END
}

/*---------------------------------------------------------------
                        getAs3DObject
  ---------------------------------------------------------------*/
void CLandmarksMap::getAs3DObject(
    mrpt::opengl::CSetOfObjects::Ptr& outObj) const
{
    if (!genericMapParams.enableSaveAs3DObject) return;

    // TODO: Generate patchs in 3D, etc...

    // Save 3D ellipsoids
    CPointPDFGaussian pointGauss;
    for (const auto& landmark : landmarks)
    {
        opengl::CEllipsoid::Ptr ellip = std::make_shared<opengl::CEllipsoid>();

        landmark.getPose(pointGauss);

        ellip->setPose(pointGauss.mean);
        ellip->setCovMatrix(pointGauss.cov);
        ellip->enableDrawSolid3D(false);
        ellip->setQuantiles(3.0);
        ellip->set3DsegmentsCount(10);
        ellip->setColor(0, 0, 1);
        ellip->setName(
            mrpt::format("LM.ID=%u", static_cast<unsigned int>(landmark.ID)));
        ellip->enableShowName(true);

        outObj->insert(ellip);
    }
}
/**** FAMD ****/
mrpt::maps::CLandmark::TLandmarkID CLandmarksMap::getMapMaxID()
{
    return _mapMaxID;
}
/**** END FAMD ****/

const CLandmark* CLandmarksMap::TCustomSequenceLandmarks::getByID(
    CLandmark::TLandmarkID ID) const
{
    for (const auto& m_landmark : m_landmarks)
    {
        if (m_landmark.ID == ID) return &m_landmark;
    }
    return nullptr;
}

// CLandmark*   CLandmarksMap::TCustomSequenceLandmarks::getByID(
// CLandmark::TLandmarkID ID )
//{
//  for(size_t indx = 0; indx < m_landmarks.size(); indx++)
//  {
//      if( m_landmarks[indx].ID == ID )
//          return &m_landmarks[indx];
//  }
//  return nullptr;
//}

const CLandmark* CLandmarksMap::TCustomSequenceLandmarks::getByBeaconID(
    unsigned int ID) const
{
    for (const auto& m_landmark : m_landmarks)
    {
        if (m_landmark.ID == ID) return &m_landmark;
    }
    return nullptr;
}

/*---------------------------------------------------------------
   Computes the ratio in [0,1] of correspondences between "this" and the
 "otherMap" map, whose 6D pose relative to "this" is "otherMapPose"
 *   In the case of a multi-metric map, this returns the average between the
 maps. This method always return 0 for grid maps.
 * \param  otherMap                   [IN] The other map to compute the matching
 with.
 * \param  otherMapPose               [IN] The 6D pose of the other map as seen
 from "this".
 * \param  maxDistForCorr             [IN] The minimum distance between 2
 non-probabilistic map elements for counting them as a correspondence.
 * \param  maxMahaDistForCorr         [IN] The minimum Mahalanobis distance
 between 2 probabilistic map elements for counting them as a correspondence.
 *
 * \return The matching ratio [0,1]
 * \sa computeMatchingWith2D
 ----------------------------------------------------------------*/
float CLandmarksMap::compute3DMatchingRatio(
    const mrpt::maps::CMetricMap* otherMap2,
    const mrpt::poses::CPose3D& otherMapPose,
    const TMatchingRatioParams& params) const
{
    MRPT_START

    // Compare to a similar map only:
    const CLandmarksMap* otherMap = nullptr;

    if (otherMap2->GetRuntimeClass() == CLASS_ID(CLandmarksMap))
        otherMap = static_cast<const CLandmarksMap*>(otherMap2);

    if (!otherMap) return 0;

    TCustomSequenceLandmarks::const_iterator itThis, itOther;
    std::deque<CPointPDFGaussian::Ptr> poses3DThis, poses3DOther;
    std::deque<CPointPDFGaussian::Ptr>::iterator itPoseThis, itPoseOther;
    CPointPDFGaussian tempPose;
    size_t nThis = landmarks.size();
    size_t nOther = otherMap->landmarks.size();
    size_t otherLandmarkWithCorrespondence = 0;

    // Checks:
    if (!nThis) return 0;
    if (!nOther) return 0;

    // The transformation:
    CMatrixDouble44 pose3DMatrix;
    otherMapPose.getHomogeneousMatrix(pose3DMatrix);
    float Tx = pose3DMatrix(0, 3);
    float Ty = pose3DMatrix(1, 3);
    float Tz = pose3DMatrix(2, 3);

    // ---------------------------------------------------------------------------------------------------------------
    // Is there any "contact" between the spheres that contain all the points
    // from each map after translating them??
    // (Note that we can avoid computing the rotation of all the points if this
    // test fail, with a great speed up!)
    // ---------------------------------------------------------------------------------------------------------------
    if (sqrt(square(Tx) + square(Ty) + square(Tz)) >=
        (landmarks.getLargestDistanceFromOrigin() +
         otherMap->landmarks.getLargestDistanceFromOrigin() + 1.0f))
        return 0;  // There is no contact!

    // Prepare:
    poses3DOther.resize(nOther);
    for (size_t i = 0; i < nOther; i++)
        poses3DOther[i] = std::make_shared<CPointPDFGaussian>();

    poses3DThis.resize(nThis);
    for (size_t i = 0; i < nThis; i++)
        poses3DThis[i] = std::make_shared<CPointPDFGaussian>();

    // Save 3D poses of the other map with transformed coordinates:
    for (itOther = otherMap->landmarks.begin(),
        itPoseOther = poses3DOther.begin();
         itOther != otherMap->landmarks.end(); itOther++, itPoseOther++)
    {
        itOther->getPose(**itPoseOther);
        (*itPoseOther)->changeCoordinatesReference(otherMapPose);
    }

    // And the 3D poses of "this" map:
    for (itThis = landmarks.begin(), itPoseThis = poses3DThis.begin();
         itThis != landmarks.end(); itThis++, itPoseThis++)
    {
        itThis->getPose(**itPoseThis);
    }

    // Check whether each "other"'s LM has a correspondence or not:
    for (itOther = otherMap->landmarks.begin(),
        itPoseOther = poses3DOther.begin();
         itOther != otherMap->landmarks.end(); itOther++, itPoseOther++)
    {
        for (itThis = landmarks.begin(), itPoseThis = poses3DThis.begin();
             itThis != landmarks.end(); itThis++, itPoseThis++)
        {
            CMatrixDouble COV =
                CMatrixDouble((*itPoseThis)->cov + (*itPoseOther)->cov);
            TPoint3D D(
                (*itPoseThis)->mean.x() - (*itPoseOther)->mean.x(),
                (*itPoseThis)->mean.y() - (*itPoseOther)->mean.y(),
                (*itPoseThis)->mean.z() - (*itPoseOther)->mean.z());
            CMatrixDouble d(1, 3);
            d(0, 0) = D.x;
            d(0, 1) = D.y;
            d(0, 2) = D.z;

            float distMaha =
                sqrt(mrpt::math::multiply_HCHt_scalar(d, COV.inverse_LLt()));

            if (distMaha < params.maxMahaDistForCorr)
            {
                // Now test the SIFT descriptors:
                if (!itThis->features.empty() && !itOther->features.empty() &&
                    itThis->features[0].descriptors.SIFT->size() ==
                        itOther->features[0].descriptors.SIFT->size())
                {
                    unsigned long descrDist = 0;
                    std::vector<unsigned char>::const_iterator it1, it2;
                    for (it1 = itThis->features[0].descriptors.SIFT->begin(),
                        it2 = itOther->features[0].descriptors.SIFT->begin();
                         it1 != itThis->features[0].descriptors.SIFT->end();
                         it1++, it2++)
                        descrDist += square(*it1 - *it2);

                    float descrDist_f =
                        sqrt(static_cast<float>(descrDist)) /
                        itThis->features[0].descriptors.SIFT->size();

                    if (descrDist_f < 1.5f)
                    {
                        otherLandmarkWithCorrespondence++;
                        break;  // Go to the next "other" LM
                    }
                }
            }
        }  // for each in "this"

    }  // for each in "other"

    return static_cast<float>(otherLandmarkWithCorrespondence) / nOther;

    MRPT_END
}

/*---------------------------------------------------------------
                    auxParticleFilterCleanUp
 ---------------------------------------------------------------*/
void CLandmarksMap::auxParticleFilterCleanUp()
{
    _mEDD.clear();
    _maxIDUpdated = false;
}

/*---------------------------------------------------------------
                    simulateRangeBearingReadings
 ---------------------------------------------------------------*/
void CLandmarksMap::simulateRangeBearingReadings(
    const CPose3D& in_robotPose, const CPose3D& in_sensorLocationOnRobot,
    CObservationBearingRange& out_Observations, bool sensorDetectsIDs,
    const float in_stdRange, const float in_stdYaw, const float in_stdPitch,
    std::vector<size_t>* out_real_associations,
    const double spurious_count_mean, const double spurious_count_std) const
{
    TSequenceLandmarks::const_iterator it;
    size_t idx;
    mrpt::obs::CObservationBearingRange::TMeasurement newMeas;
    CPoint3D beacon3D;
    CPointPDFGaussian beaconPDF;

    if (out_real_associations) out_real_associations->clear();

    // Compute the 3D position of the sensor:
    const CPose3D point3D = in_robotPose + CPose3D(in_sensorLocationOnRobot);

    // Clear output data:
    out_Observations.validCovariances = false;
    out_Observations.sensor_std_range = in_stdRange;
    out_Observations.sensor_std_yaw = in_stdYaw;
    out_Observations.sensor_std_pitch = in_stdPitch;

    out_Observations.sensedData.clear();
    out_Observations.timestamp = mrpt::system::getCurrentTime();
    out_Observations.sensorLocationOnRobot = in_sensorLocationOnRobot;

    // For each BEACON landmark in the map:
    // ------------------------------------------
    for (idx = 0, it = landmarks.begin(); it != landmarks.end(); ++it, ++idx)
    {
        // Get the 3D position of the beacon (just the mean value):
        it->getPose(beaconPDF);
        beacon3D = CPoint3D(beaconPDF.mean);

        // Compute yaw,pitch,range:
        double range, yaw, pitch;
        point3D.sphericalCoordinates(beacon3D.asTPoint(), range, yaw, pitch);

        // Add noises:
        range += in_stdRange * getRandomGenerator().drawGaussian1D_normalized();
        yaw += in_stdYaw * getRandomGenerator().drawGaussian1D_normalized();
        pitch += in_stdPitch * getRandomGenerator().drawGaussian1D_normalized();

        yaw = math::wrapToPi(yaw);
        range = max(0.0, range);

        if (range >= out_Observations.minSensorDistance &&
            range <= out_Observations.maxSensorDistance &&
            fabs(yaw) <= 0.5f * out_Observations.fieldOfView_yaw &&
            fabs(pitch) <= 0.5f * out_Observations.fieldOfView_pitch)
        {
            // Fill out:
            if (sensorDetectsIDs)
                newMeas.landmarkID = it->ID;
            else
                newMeas.landmarkID = INVALID_LANDMARK_ID;
            newMeas.range = range;
            newMeas.yaw = yaw;
            newMeas.pitch = pitch;

            // Insert:
            out_Observations.sensedData.push_back(newMeas);

            if (out_real_associations)
                out_real_associations->push_back(idx);  // Real indices.
        }
    }  // end for it

    const double fSpurious = getRandomGenerator().drawGaussian1D(
        spurious_count_mean, spurious_count_std);
    size_t nSpurious = 0;
    if (spurious_count_std != 0 || spurious_count_mean != 0)
        nSpurious =
            static_cast<size_t>(mrpt::round_long(std::max(0.0, fSpurious)));

    // For each spurios reading to generate:
    // ------------------------------------------
    for (size_t i = 0; i < nSpurious; i++)
    {
        // Make up yaw,pitch,range out from thin air:
        // (the conditionals on yaw & pitch are to generate 2D observations if
        // we are in 2D, which we learn from a null std.dev.)
        const double range = getRandomGenerator().drawUniform(
            out_Observations.minSensorDistance,
            out_Observations.maxSensorDistance);
        const double yaw = (out_Observations.sensor_std_yaw == 0)
                               ? 0
                               : getRandomGenerator().drawUniform(
                                     -0.5f * out_Observations.fieldOfView_yaw,
                                     0.5f * out_Observations.fieldOfView_yaw);
        const double pitch =
            (out_Observations.sensor_std_pitch == 0)
                ? 0
                : getRandomGenerator().drawUniform(
                      -0.5f * out_Observations.fieldOfView_pitch,
                      0.5f * out_Observations.fieldOfView_pitch);

        // Fill out:
        newMeas.landmarkID =
            INVALID_LANDMARK_ID;  // Always invalid ID since it's spurious
        newMeas.range = range;
        newMeas.yaw = yaw;
        newMeas.pitch = pitch;

        // Insert:
        out_Observations.sensedData.push_back(newMeas);

        if (out_real_associations)
            out_real_associations->push_back(
                std::string::npos);  // Real index: spurious
    }  // end for it

    // Done!
}