tracking.cpp

Go to the documentation of this file.

/* +------------------------------------------------------------------------+
   |                     Mobile Robot Programming Toolkit (MRPT)            |
   |                          http://www.mrpt.org/                          |
   |                                                                        |
   | Copyright (c) 2005-2018, Individual contributors, see AUTHORS file     |
   | See: http://www.mrpt.org/Authors - All rights reserved.                |
   | Released under BSD License. See details in http://www.mrpt.org/License |
   +------------------------------------------------------------------------+ */

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

#include <mrpt/vision/tracking.h>
#include <mrpt/vision/CFeatureExtraction.h>

// Universal include for all versions of OpenCV
#include <mrpt/otherlibs/do_opencv_includes.h>

using namespace mrpt;
using namespace mrpt::vision;
using namespace mrpt::img;
using namespace mrpt::tfest;
using namespace mrpt::math;
using namespace std;

// ------------------------------- internal helper templates
// ---------------------------------
namespace mrpt::vision::detail
{

template <typename FEATLIST>
inline void trackFeatures_checkResponses(
    FEATLIST& featureList, const CImage& cur_gray,
    const float minimum_KLT_response, const unsigned int KLT_response_half_win,
    const unsigned int max_x, const unsigned int max_y);

template <>
inline void trackFeatures_checkResponses<CFeatureList>(
    CFeatureList& featureList, const CImage& cur_gray,
    const float minimum_KLT_response, const unsigned int KLT_response_half_win,
    const unsigned int max_x, const unsigned int max_y)
{
    const CFeatureList::iterator itFeatEnd = featureList.end();
    for (CFeatureList::iterator itFeat = featureList.begin();
         itFeat != itFeatEnd; ++itFeat)
    {
        CFeature* ft = itFeat->get();
        if (ft->track_status != status_TRACKED)
            continue;  // Skip if it's not correctly tracked.

        const unsigned int x = ft->x;
        const unsigned int y = ft->y;
        if (x > KLT_response_half_win && y > KLT_response_half_win &&
            x < max_x && y < max_y)
        {  // Update response:
            ft->response = cur_gray.KLT_response(x, y, KLT_response_half_win);

            // Is it good enough?
            // http://grooveshark.com/s/Goonies+Are+Good+Enough/2beBfO?src=5
            if (ft->response < minimum_KLT_response)
            {  // Nope!
                ft->track_status = status_LOST;
            }
        }
        else
        {  // Out of bounds
            ft->response = 0;
            ft->track_status = status_OOB;
        }
    }
}  // end of trackFeatures_checkResponses<>

template <class FEAT_LIST>
inline void trackFeatures_checkResponses_impl_simple(
    FEAT_LIST& featureList, const CImage& cur_gray,
    const float minimum_KLT_response, const unsigned int KLT_response_half_win,
    const unsigned int max_x_, const unsigned int max_y_)
{
    if (featureList.empty()) return;

    using pixel_coord_t = typename FEAT_LIST::feature_t::pixel_coord_t;
    const pixel_coord_t half_win =
        static_cast<pixel_coord_t>(KLT_response_half_win);
    const pixel_coord_t max_x = static_cast<pixel_coord_t>(max_x_);
    const pixel_coord_t max_y = static_cast<pixel_coord_t>(max_y_);

    for (int N = featureList.size() - 1; N >= 0; --N)
    {
        typename FEAT_LIST::feature_t& ft = featureList[N];
        if (ft.track_status != status_TRACKED)
            continue;  // Skip if it's not correctly tracked.

        if (ft.pt.x > half_win && ft.pt.y > half_win && ft.pt.x < max_x &&
            ft.pt.y < max_y)
        {  // Update response:
            ft.response =
                cur_gray.KLT_response(ft.pt.x, ft.pt.y, KLT_response_half_win);

            // Is it good enough?
            // http://grooveshark.com/s/Goonies+Are+Good+Enough/2beBfO?src=5
            if (ft.response < minimum_KLT_response)
            {  // Nope!
                ft.track_status = status_LOST;
            }
        }
        else
        {  // Out of bounds
            ft.response = 0;
            ft.track_status = status_OOB;
        }
    }
}  // end of trackFeatures_checkResponses<>

template <>
inline void trackFeatures_checkResponses<TSimpleFeatureList>(
    TSimpleFeatureList& featureList, const CImage& cur_gray,
    const float minimum_KLT_response, const unsigned int KLT_response_half_win,
    const unsigned int max_x, const unsigned int max_y)
{
    trackFeatures_checkResponses_impl_simple<TSimpleFeatureList>(
        featureList, cur_gray, minimum_KLT_response, KLT_response_half_win,
        max_x, max_y);
}
template <>
inline void trackFeatures_checkResponses<TSimpleFeaturefList>(
    TSimpleFeaturefList& featureList, const CImage& cur_gray,
    const float minimum_KLT_response, const unsigned int KLT_response_half_win,
    const unsigned int max_x, const unsigned int max_y)
{
    trackFeatures_checkResponses_impl_simple<TSimpleFeaturefList>(
        featureList, cur_gray, minimum_KLT_response, KLT_response_half_win,
        max_x, max_y);
}

template <typename FEATLIST>
inline void trackFeatures_updatePatch(
    FEATLIST& featureList, const CImage& cur_gray);

template <>
inline void trackFeatures_updatePatch<CFeatureList>(
    CFeatureList& featureList, const CImage& cur_gray)
{
    for (CFeatureList::iterator itFeat = featureList.begin();
         itFeat != featureList.end(); ++itFeat)
    {
        CFeature* ft = itFeat->get();
        if (ft->track_status != status_TRACKED)
            continue;  // Skip if it's not correctly tracked.

        const size_t patch_width = ft->patch.getWidth();
        const size_t patch_height = ft->patch.getHeight();
        if (patch_width > 0 && patch_height > 0)
        {
            try
            {
                const int offset = (int)patch_width / 2;  // + 1;
                cur_gray.extract_patch(
                    ft->patch, round(ft->x) - offset, round(ft->y) - offset,
                    patch_width, patch_height);
            }
            catch (std::exception&)
            {
                ft->track_status = status_OOB;  // Out of bounds!
            }
        }
    }
}  // end of trackFeatures_updatePatch<>
template <>
inline void trackFeatures_updatePatch<TSimpleFeatureList>(
    TSimpleFeatureList& featureList, const CImage& cur_gray)
{
    MRPT_UNUSED_PARAM(featureList);
    MRPT_UNUSED_PARAM(cur_gray);
    // This list type does not have patch stored explicitly
}  // end of trackFeatures_updatePatch<>
template <>
inline void trackFeatures_updatePatch<TSimpleFeaturefList>(
    TSimpleFeaturefList& featureList, const CImage& cur_gray)
{
    MRPT_UNUSED_PARAM(featureList);
    MRPT_UNUSED_PARAM(cur_gray);
    // This list type does not have patch stored explicitly
}  // end of trackFeatures_updatePatch<>

template <typename FEATLIST>
inline void trackFeatures_addNewFeats(
    FEATLIST& featureList, const TSimpleFeatureList& new_feats,
    const std::vector<size_t>& sorted_indices, const size_t nNewToCheck,
    const size_t maxNumFeatures, const float minimum_KLT_response_to_add,
    const double threshold_sqr_dist_to_add_new, const size_t patchSize,
    const CImage& cur_gray, TFeatureID& max_feat_ID_at_input);

template <>
inline void trackFeatures_addNewFeats<CFeatureList>(
    CFeatureList& featureList, const TSimpleFeatureList& new_feats,
    const std::vector<size_t>& sorted_indices, const size_t nNewToCheck,
    const size_t maxNumFeatures, const float minimum_KLT_response_to_add,
    const double threshold_sqr_dist_to_add_new, const size_t patchSize,
    const CImage& cur_gray, TFeatureID& max_feat_ID_at_input)
{
    const TImageSize imgSize = cur_gray.getSize();
    const int offset = (int)patchSize / 2 + 1;
    const int w_off = int(imgSize.x - offset);
    const int h_off = int(imgSize.y - offset);

    for (size_t i = 0; i < nNewToCheck && featureList.size() < maxNumFeatures;
         i++)
    {
        const TSimpleFeature& feat = new_feats[sorted_indices[i]];

        if (feat.response < minimum_KLT_response_to_add) continue;

        double min_dist_sqr = square(10000);

        if (!featureList.empty())
        {
            // m_timlog.enter("[CGenericFeatureTracker] add new
            // features.kdtree");
            min_dist_sqr =
                featureList.kdTreeClosestPoint2DsqrError(feat.pt.x, feat.pt.y);
            // m_timlog.leave("[CGenericFeatureTracker] add new
            // features.kdtree");
        }

        if (min_dist_sqr > threshold_sqr_dist_to_add_new &&
            feat.pt.x > offset && feat.pt.y > offset && feat.pt.x < w_off &&
            feat.pt.y < h_off)
        {
            // Add new feature:
            CFeature::Ptr ft = mrpt::make_aligned_shared<CFeature>();
            ft->type = featFAST;
            ft->ID = ++max_feat_ID_at_input;
            ft->x = feat.pt.x;
            ft->y = feat.pt.y;
            ft->response = feat.response;
            ft->orientation = 0;
            ft->scale = 1;
            ft->patchSize = patchSize;  // The size of the feature patch

            if (patchSize > 0)
                cur_gray.extract_patch(
                    ft->patch, round(ft->x) - offset, round(ft->y) - offset,
                    patchSize,
                    patchSize);  // Image patch surronding the feature

            featureList.push_back(ft);
        }
    }
}  // end of trackFeatures_addNewFeats<>

template <class FEAT_LIST>
inline void trackFeatures_addNewFeats_simple_list(
    FEAT_LIST& featureList, const TSimpleFeatureList& new_feats,
    const std::vector<size_t>& sorted_indices, const size_t nNewToCheck,
    const size_t maxNumFeatures, const float minimum_KLT_response_to_add,
    const double threshold_sqr_dist_to_add_new, const size_t patchSize,
    const CImage& cur_gray, TFeatureID& max_feat_ID_at_input)
{
#if 0
                // Brute-force version:
                const int max_manhatan_dist = std::sqrt(2*threshold_sqr_dist_to_add_new);

                for (size_t i=0;i<nNewToCheck && featureList.size()<maxNumFeatures;i++)
                {
                    const TSimpleFeature &feat = new_feats[ sorted_indices[i] ];
                    if (feat.response<minimum_KLT_response_to_add) break; // continue;

                    // Check the min-distance:
                    int manh_dist = std::numeric_limits<int>::max();
                    for (size_t j=0;j<featureList.size();j++)
                    {
                        const TSimpleFeature &existing = featureList[j];
                        const int d = std::abs(existing.pt.x-feat.pt.x)+std::abs(existing.pt.y-feat.pt.y);
                        mrpt::keep_min(manh_dist, d);
                    }

                    if (manh_dist<max_manhatan_dist)
                        continue; // Already occupied! skip.

                    // OK: accept it
                    featureList.push_back_fast(feat.pt.x,feat.pt.y);  // (x,y)
                    //featureList.mark_kdtree_as_outdated();

                    // Fill out the rest of data:
                    TSimpleFeature &newFeat = featureList.back();

                    newFeat.ID          = ++max_feat_ID_at_input;
                    newFeat.response    = feat.response;
                    newFeat.octave      = 0;
                    /** Inactive: right after detection, and before being tried to track */
                    newFeat.track_status = status_IDLE;
                }
#elif 0
    // Version with an occupancy grid:
    const int grid_cell_log2 = round(
        std::log(std::sqrt(threshold_sqr_dist_to_add_new) * 0.5) /
        std::log(2.0));

    int grid_lx = 1 + (cur_gray.getWidth() >> grid_cell_log2);
    int grid_ly = 1 + (cur_gray.getHeight() >> grid_cell_log2);

    mrpt::math::CMatrixBool& occupied_sections =
        featureList.getOccupiedSectionsMatrix();

    occupied_sections.setSize(
        grid_lx, grid_ly);  // See the comments above for an explanation.
    occupied_sections.fillAll(false);

    for (size_t i = 0; i < featureList.size(); i++)
    {
        const TSimpleFeature& feat = featureList[i];
        const int section_idx_x = feat.pt.x >> grid_cell_log2;
        const int section_idx_y = feat.pt.y >> grid_cell_log2;

        if (!section_idx_x || !section_idx_y || section_idx_x >= grid_lx - 1 ||
            section_idx_y >= grid_ly - 1)
            continue;  // This may be too radical, but speeds up the logic
        // below...

        // Mark sections as occupied
        bool* ptr1 =
            &occupied_sections.get_unsafe(section_idx_x - 1, section_idx_y - 1);
        bool* ptr2 =
            &occupied_sections.get_unsafe(section_idx_x - 1, section_idx_y);
        bool* ptr3 =
            &occupied_sections.get_unsafe(section_idx_x - 1, section_idx_y + 1);
        ptr1[0] = ptr1[1] = ptr1[2] = true;
        ptr2[0] = ptr2[1] = ptr2[2] = true;
        ptr3[0] = ptr3[1] = ptr3[2] = true;
    }

    for (size_t i = 0; i < nNewToCheck && featureList.size() < maxNumFeatures;
         i++)
    {
        const TSimpleFeature& feat = new_feats[sorted_indices[i]];
        if (feat.response < minimum_KLT_response_to_add) break;  // continue;

        // Check the min-distance:
        const int section_idx_x = feat.pt.x >> grid_cell_log2;
        const int section_idx_y = feat.pt.y >> grid_cell_log2;

        if (!section_idx_x || !section_idx_y || section_idx_x >= grid_lx - 2 ||
            section_idx_y >= grid_ly - 2)
            continue;  // This may be too radical, but speeds up the logic
        // below...

        if (occupied_sections(section_idx_x, section_idx_y))
            continue;  // Already occupied! skip.

        // Mark section as occupied
        bool* ptr1 =
            &occupied_sections.get_unsafe(section_idx_x - 1, section_idx_y - 1);
        bool* ptr2 =
            &occupied_sections.get_unsafe(section_idx_x - 1, section_idx_y);
        bool* ptr3 =
            &occupied_sections.get_unsafe(section_idx_x - 1, section_idx_y + 1);

        ptr1[0] = ptr1[1] = ptr1[2] = true;
        ptr2[0] = ptr2[1] = ptr2[2] = true;
        ptr3[0] = ptr3[1] = ptr3[2] = true;

        // OK: accept it
        featureList.push_back_fast(feat.pt.x, feat.pt.y);  // (x,y)
        // featureList.mark_kdtree_as_outdated();

        // Fill out the rest of data:
        TSimpleFeature& newFeat = featureList.back();

        newFeat.ID = ++max_feat_ID_at_input;
        newFeat.response = feat.response;
        newFeat.octave = 0;
        /** Inactive: right after detection, and before being tried to track */
        newFeat.track_status = status_IDLE;
    }
#else
    MRPT_UNUSED_PARAM(patchSize);
    MRPT_UNUSED_PARAM(cur_gray);
    // Version with KD-tree
    CFeatureListKDTree<typename FEAT_LIST::feature_t> kdtree(
        featureList.getVector());

    for (size_t i = 0; i < nNewToCheck && featureList.size() < maxNumFeatures;
         i++)
    {
        const TSimpleFeature& feat = new_feats[sorted_indices[i]];
        if (feat.response < minimum_KLT_response_to_add) break;  // continue;

        // Check the min-distance:
        double min_dist_sqr = std::numeric_limits<double>::max();

        if (!featureList.empty())
        {
            // m_timlog.enter("[CGenericFeatureTracker] add new
            // features.kdtree");
            min_dist_sqr =
                kdtree.kdTreeClosestPoint2DsqrError(feat.pt.x, feat.pt.y);
            // m_timlog.leave("[CGenericFeatureTracker] add new
            // features.kdtree");
        }

        if (min_dist_sqr > threshold_sqr_dist_to_add_new)
        {
            // OK: accept it
            featureList.push_back_fast(feat.pt.x, feat.pt.y);  // (x,y)
            kdtree.mark_as_outdated();

            // Fill out the rest of data:
            typename FEAT_LIST::feature_t& newFeat = featureList.back();

            newFeat.ID = ++max_feat_ID_at_input;
            newFeat.response = feat.response;
            newFeat.octave = 0;
            /** Inactive: right after detection, and before being tried to track
             */
            newFeat.track_status = status_IDLE;
        }
    }

#endif
}  // end of trackFeatures_addNewFeats<>

template <>
inline void trackFeatures_addNewFeats<TSimpleFeatureList>(
    TSimpleFeatureList& featureList, const TSimpleFeatureList& new_feats,
    const std::vector<size_t>& sorted_indices, const size_t nNewToCheck,
    const size_t maxNumFeatures, const float minimum_KLT_response_to_add,
    const double threshold_sqr_dist_to_add_new, const size_t patchSize,
    const CImage& cur_gray, TFeatureID& max_feat_ID_at_input)
{
    trackFeatures_addNewFeats_simple_list<TSimpleFeatureList>(
        featureList, new_feats, sorted_indices, nNewToCheck, maxNumFeatures,
        minimum_KLT_response_to_add, threshold_sqr_dist_to_add_new, patchSize,
        cur_gray, max_feat_ID_at_input);
}
template <>
inline void trackFeatures_addNewFeats<TSimpleFeaturefList>(
    TSimpleFeaturefList& featureList, const TSimpleFeatureList& new_feats,
    const std::vector<size_t>& sorted_indices, const size_t nNewToCheck,
    const size_t maxNumFeatures, const float minimum_KLT_response_to_add,
    const double threshold_sqr_dist_to_add_new, const size_t patchSize,
    const CImage& cur_gray, TFeatureID& max_feat_ID_at_input)
{
    trackFeatures_addNewFeats_simple_list<TSimpleFeaturefList>(
        featureList, new_feats, sorted_indices, nNewToCheck, maxNumFeatures,
        minimum_KLT_response_to_add, threshold_sqr_dist_to_add_new, patchSize,
        cur_gray, max_feat_ID_at_input);
}

// Return the number of removed features
template <typename FEATLIST>
inline size_t trackFeatures_deleteOOB(
    FEATLIST& trackedFeats, const size_t img_width, const size_t img_height,
    const int MIN_DIST_MARGIN_TO_STOP_TRACKING);

template <typename FEATLIST>
inline size_t trackFeatures_deleteOOB_impl_simple_feat(
    FEATLIST& trackedFeats, const size_t img_width, const size_t img_height,
    const int MIN_DIST_MARGIN_TO_STOP_TRACKING)
{
    if (trackedFeats.empty()) return 0;

    std::vector<size_t> survival_idxs;
    const size_t N = trackedFeats.size();

    // 1st: Build list of survival indexes:
    survival_idxs.reserve(N);
    for (size_t i = 0; i < N; i++)
    {
        const typename FEATLIST::feature_t& ft = trackedFeats[i];
        const TFeatureTrackStatus status = ft.track_status;
        bool eras = (status_TRACKED != status && status_IDLE != status);
        if (!eras)
        {
            // Also, check if it's too close to the image border:
            const int x = ft.pt.x;
            const int y = ft.pt.y;
            if (x < MIN_DIST_MARGIN_TO_STOP_TRACKING ||
                y < MIN_DIST_MARGIN_TO_STOP_TRACKING ||
                x > static_cast<int>(
                        img_width - MIN_DIST_MARGIN_TO_STOP_TRACKING) ||
                y > static_cast<int>(
                        img_height - MIN_DIST_MARGIN_TO_STOP_TRACKING))
            {
                eras = true;
            }
        }
        if (!eras) survival_idxs.push_back(i);
    }

    // 2nd: Build updated list:
    const size_t N2 = survival_idxs.size();
    const size_t n_removed = N - N2;
    for (size_t i = 0; i < N2; i++)
    {
        if (survival_idxs[i] != i)
            trackedFeats[i] = trackedFeats[survival_idxs[i]];
    }
    trackedFeats.resize(N2);
    return n_removed;
}  // end of trackFeatures_deleteOOB

template <>
inline size_t trackFeatures_deleteOOB(
    TSimpleFeatureList& trackedFeats, const size_t img_width,
    const size_t img_height, const int MIN_DIST_MARGIN_TO_STOP_TRACKING)
{
    return trackFeatures_deleteOOB_impl_simple_feat<TSimpleFeatureList>(
        trackedFeats, img_width, img_height, MIN_DIST_MARGIN_TO_STOP_TRACKING);
}
template <>
inline size_t trackFeatures_deleteOOB(
    TSimpleFeaturefList& trackedFeats, const size_t img_width,
    const size_t img_height, const int MIN_DIST_MARGIN_TO_STOP_TRACKING)
{
    return trackFeatures_deleteOOB_impl_simple_feat<TSimpleFeaturefList>(
        trackedFeats, img_width, img_height, MIN_DIST_MARGIN_TO_STOP_TRACKING);
}

template <>
inline size_t trackFeatures_deleteOOB(
    CFeatureList& trackedFeats, const size_t img_width, const size_t img_height,
    const int MIN_DIST_MARGIN_TO_STOP_TRACKING)
{
    CFeatureList::iterator itFeat = trackedFeats.begin();
    size_t n_removed = 0;
    while (itFeat != trackedFeats.end())
    {
        const TFeatureTrackStatus status = (*itFeat)->track_status;
        bool eras = (status_TRACKED != status && status_IDLE != status);
        if (!eras)
        {
            // Also, check if it's too close to the image border:
            const float x = (*itFeat)->x;
            const float y = (*itFeat)->y;
            if (x < MIN_DIST_MARGIN_TO_STOP_TRACKING ||
                y < MIN_DIST_MARGIN_TO_STOP_TRACKING ||
                x > (img_width - MIN_DIST_MARGIN_TO_STOP_TRACKING) ||
                y > (img_height - MIN_DIST_MARGIN_TO_STOP_TRACKING))
            {
                eras = true;
            }
        }
        if (eras)  // Erase or keep?
        {
            itFeat = trackedFeats.erase(itFeat);
            n_removed++;
        }
        else
            ++itFeat;
    }
    return n_removed;
}  // end of trackFeatures_deleteOOB
}  // end NS's
// ---------------------------- end of internal helper templates
// -------------------------------

void CGenericFeatureTracker::trackFeatures_impl(
    const CImage& old_img, const CImage& new_img,
    TSimpleFeaturefList& inout_featureList)
{
    MRPT_UNUSED_PARAM(old_img);
    MRPT_UNUSED_PARAM(new_img);
    MRPT_UNUSED_PARAM(inout_featureList);
    THROW_EXCEPTION("Method not implemented by derived class!");
}

/** Perform feature tracking from "old_img" to "new_img", with a (possibly
  *empty) list of previously tracked features "featureList".
  *  This is a list of parameters (in "extraParams") accepted by ALL
  *implementations of feature tracker (see each derived class for more specific
  *parameters).
  *     - "add_new_features" (Default=0). If set to "1", new features will be
  *also
  *added to the existing ones in areas of the image poor of features.
  * This method actually first call the pure virtual "trackFeatures_impl"
  *method, then implements the optional detection of new features if
  *"add_new_features"!=0.
  */
template <typename FEATLIST>
void CGenericFeatureTracker::internal_trackFeatures(
    const CImage& old_img, const CImage& new_img, FEATLIST& featureList)
{
    m_timlog.enter(
        "[CGenericFeatureTracker::trackFeatures] Complete iteration");

    const size_t img_width = new_img.getWidth();
    const size_t img_height = new_img.getHeight();

    // Take the maximum ID of "old" features so new feats (if
    // "add_new_features==true") will be id+1, id+2, ...
    TFeatureID max_feat_ID_at_input = 0;
    if (!featureList.empty()) max_feat_ID_at_input = featureList.getMaxID();

    // Grayscale images
    // =========================================
    m_timlog.enter("[CGenericFeatureTracker] Convert grayscale");

    const CImage prev_gray(old_img, FAST_REF_OR_CONVERT_TO_GRAY);
    const CImage cur_gray(new_img, FAST_REF_OR_CONVERT_TO_GRAY);

    m_timlog.leave("[CGenericFeatureTracker] Convert grayscale");

    // =================================
    // (1st STEP)  Do the actual tracking
    // =================================
    m_newly_detected_feats.clear();

    m_timlog.enter("[CGenericFeatureTracker] trackFeatures_impl");

    trackFeatures_impl(prev_gray, cur_gray, featureList);

    m_timlog.leave("[CGenericFeatureTracker] trackFeatures_impl");

    // ========================================================
    // (2nd STEP) For successfully followed features, check their KLT response??
    // ========================================================
    const int check_KLT_response_every =
        extra_params.getWithDefaultVal("check_KLT_response_every", 0);
    const float minimum_KLT_response =
        extra_params.getWithDefaultVal("minimum_KLT_response", 5);
    const unsigned int KLT_response_half_win =
        extra_params.getWithDefaultVal("KLT_response_half_win", 4);

    if (check_KLT_response_every > 0 &&
        ++m_check_KLT_counter >= size_t(check_KLT_response_every))
    {
        m_timlog.enter("[CGenericFeatureTracker] check KLT responses");
        m_check_KLT_counter = 0;

        const unsigned int max_x = img_width - KLT_response_half_win;
        const unsigned int max_y = img_height - KLT_response_half_win;

        detail::trackFeatures_checkResponses(
            featureList, cur_gray, minimum_KLT_response, KLT_response_half_win,
            max_x, max_y);

        m_timlog.leave("[CGenericFeatureTracker] check KLT responses");

    }  // end check_KLT_response_every

    // ============================================================
    // (3rd STEP)  Remove Out-of-bounds or badly tracked features
    //   or those marked as "bad" by their low KLT response
    // ============================================================
    const bool remove_lost_features =
        extra_params.getWithDefaultVal("remove_lost_features", 0) != 0;

    if (remove_lost_features)
    {
        m_timlog.enter("[CGenericFeatureTracker] removal of OOB");

        static const int MIN_DIST_MARGIN_TO_STOP_TRACKING = 10;

        const size_t nRemoved = detail::trackFeatures_deleteOOB(
            featureList, img_width, img_height,
            MIN_DIST_MARGIN_TO_STOP_TRACKING);

        m_timlog.leave("[CGenericFeatureTracker] removal of OOB");

        last_execution_extra_info.num_deleted_feats = nRemoved;
    }
    else
    {
        last_execution_extra_info.num_deleted_feats = 0;
    }

    // ========================================================
    // (4th STEP) For successfully followed features, update its patch:
    // ========================================================
    const int update_patches_every =
        extra_params.getWithDefaultVal("update_patches_every", 0);

    if (update_patches_every > 0 &&
        ++m_update_patches_counter >= size_t(update_patches_every))
    {
        m_timlog.enter("[CGenericFeatureTracker] update patches");
        m_update_patches_counter = 0;

        // Update the patch for each valid feature:
        detail::trackFeatures_updatePatch(featureList, cur_gray);

        m_timlog.leave("[CGenericFeatureTracker] update patches");
    }  // end if update_patches_every

    // ========================================================
    // (5th STEP) Do detection of new features??
    // ========================================================
    const bool add_new_features =
        extra_params.getWithDefaultVal("add_new_features", 0) != 0;
    const double threshold_dist_to_add_new =
        extra_params.getWithDefaultVal("add_new_feat_min_separation", 15);

    // Additional operation: if "add_new_features==true", find new features and
    // add them in
    //  areas spare of valid features:
    if (add_new_features)
    {
        m_timlog.enter("[CGenericFeatureTracker] add new features");

        // Look for new features and save in "m_newly_detected_feats", if
        // they're not already computed:
        if (m_newly_detected_feats.empty())
        {
            // Do the detection
            CFeatureExtraction::detectFeatures_SSE2_FASTER12(
                cur_gray, m_newly_detected_feats, m_detector_adaptive_thres);
        }

        const size_t N = m_newly_detected_feats.size();

        last_execution_extra_info.raw_FAST_feats_detected =
            N;  // Extra out info.

        // Update the adaptive threshold.
        const size_t desired_num_features = extra_params.getWithDefaultVal(
            "desired_num_features_adapt",
            size_t((img_width * img_height) >> 9));
        updateAdaptiveNewFeatsThreshold(N, desired_num_features);

        // Use KLT response instead of the OpenCV's original "response" field:
        {
            const unsigned int max_x = img_width - KLT_response_half_win;
            const unsigned int max_y = img_height - KLT_response_half_win;
            for (size_t i = 0; i < N; i++)
            {
                const unsigned int x = m_newly_detected_feats[i].pt.x;
                const unsigned int y = m_newly_detected_feats[i].pt.y;
                if (x > KLT_response_half_win && y > KLT_response_half_win &&
                    x < max_x && y < max_y)
                    m_newly_detected_feats[i].response =
                        cur_gray.KLT_response(x, y, KLT_response_half_win);
                else
                    m_newly_detected_feats[i].response = 0;  // Out of bounds
            }
        }

        //  Sort them by "response": It's ~100 times faster to sort a list of
        //      indices "sorted_indices" than sorting directly the actual list
        //      of features "m_newly_detected_feats"
        std::vector<size_t> sorted_indices(N);
        for (size_t i = 0; i < N; i++) sorted_indices[i] = i;

        std::sort(
            sorted_indices.begin(), sorted_indices.end(),
            KeypointResponseSorter<TSimpleFeatureList>(m_newly_detected_feats));

        // For each new good feature, add it to the list of tracked ones only if
        // it's pretty
        //  isolated:

        const size_t nNewToCheck = std::min(size_t(1500), N);
        const double threshold_sqr_dist_to_add_new =
            square(threshold_dist_to_add_new);
        const size_t maxNumFeatures =
            extra_params.getWithDefaultVal("add_new_feat_max_features", 100);
        const size_t patchSize =
            extra_params.getWithDefaultVal("add_new_feat_patch_size", 11);

        const float minimum_KLT_response_to_add =
            extra_params.getWithDefaultVal("minimum_KLT_response_to_add", 10);

        // Do it:
        detail::trackFeatures_addNewFeats(
            featureList, m_newly_detected_feats, sorted_indices, nNewToCheck,
            maxNumFeatures, minimum_KLT_response_to_add,
            threshold_sqr_dist_to_add_new, patchSize, cur_gray,
            max_feat_ID_at_input);

        m_timlog.leave("[CGenericFeatureTracker] add new features");
    }

    m_timlog.leave(
        "[CGenericFeatureTracker::trackFeatures] Complete iteration");

}  // end of CGenericFeatureTracker::trackFeatures

void CGenericFeatureTracker::trackFeatures(
    const CImage& old_img, const CImage& new_img, CFeatureList& featureList)
{
    internal_trackFeatures<CFeatureList>(old_img, new_img, featureList);
}

void CGenericFeatureTracker::trackFeatures(
    const CImage& old_img, const CImage& new_img,
    TSimpleFeatureList& featureList)
{
    internal_trackFeatures<TSimpleFeatureList>(old_img, new_img, featureList);
}

void CGenericFeatureTracker::trackFeatures(
    const CImage& old_img, const CImage& new_img,
    TSimpleFeaturefList& featureList)
{
    internal_trackFeatures<TSimpleFeaturefList>(old_img, new_img, featureList);
}

void CGenericFeatureTracker::updateAdaptiveNewFeatsThreshold(
    const size_t nNewlyDetectedFeats, const size_t desired_num_features)
{
    const size_t hysteresis_min_num_feats = desired_num_features * 0.9;
    const size_t hysteresis_max_num_feats = desired_num_features * 1.1;

    if (nNewlyDetectedFeats < hysteresis_min_num_feats)
        m_detector_adaptive_thres = std::max(
            2.0, std::min(
                     m_detector_adaptive_thres - 1.0,
                     m_detector_adaptive_thres * 0.8));
    else if (nNewlyDetectedFeats > hysteresis_max_num_feats)
        m_detector_adaptive_thres = std::max(
            m_detector_adaptive_thres + 1.0, m_detector_adaptive_thres * 1.2);
}

/*------------------------------------------------------------
                    checkTrackedFeatures
-------------------------------------------------------------*/
void vision::checkTrackedFeatures(
    CFeatureList& leftList, CFeatureList& rightList,
    vision::TMatchingOptions options)
{
    ASSERT_(leftList.size() == rightList.size());

    // std::cout << std::endl << "Tracked features checking ..." << std::endl;

    CFeatureList::iterator itLeft, itRight;
    size_t u, v;
    double res;

    for (itLeft = leftList.begin(), itRight = rightList.begin();
         itLeft != leftList.end();)
    {
        bool delFeat = false;
        if ((*itLeft)->x < 0 || (*itLeft)->y < 0 ||  // Out of bounds
            (*itRight)->x < 0 || (*itRight)->y < 0 ||  // Out of bounds
            fabs((*itLeft)->y - (*itRight)->y) >
                options
                    .epipolar_TH)  // Not fulfillment of the epipolar constraint
        {
            // Show reason
            std::cout << "Bad tracked match:";
            if ((*itLeft)->x < 0 || (*itLeft)->y < 0 || (*itRight)->x < 0 ||
                (*itRight)->y < 0)
                std::cout << " Out of bounds: (" << (*itLeft)->x << ","
                          << (*itLeft)->y << " & (" << (*itRight)->x << ","
                          << (*itRight)->y << ")" << std::endl;

            if (fabs((*itLeft)->y - (*itRight)->y) > options.epipolar_TH)
                std::cout << " Bad row checking: "
                          << fabs((*itLeft)->y - (*itRight)->y) << std::endl;

            delFeat = true;
        }
        else
        {
            // Compute cross correlation:
            openCV_cross_correlation(
                (*itLeft)->patch, (*itRight)->patch, u, v, res);

            if (res < options.minCC_TH)
            {
                std::cout << "Bad tracked match (correlation failed):"
                          << " CC Value: " << res << std::endl;
                delFeat = true;
            }
        }  // end if

        if (delFeat)  // Erase the pair of features
        {
            itLeft = leftList.erase(itLeft);
            itRight = rightList.erase(itRight);
        }
        else
        {
            itLeft++;
            itRight++;
        }
    }  // end for
}  // end checkTrackedFeatures

/*-------------------------------------------------------------
                    filterBadCorrsByDistance
-------------------------------------------------------------*/
void vision::filterBadCorrsByDistance(
    TMatchingPairList& feat_list, unsigned int numberOfSigmas)
{
    ASSERT_(numberOfSigmas > 0);
    //  MRPT_UNUSED_PARAM( numberOfSigmas );
    MRPT_START

    TMatchingPairList::iterator itPair;
    CMatrix dist;
    double v_mean, v_std;
    unsigned int count = 0;

    dist.setSize(feat_list.size(), 1);
    // v_mean.resize(1);
    // v_std.resize(1);

    // Compute mean and standard deviation of the distance
    for (itPair = feat_list.begin(); itPair != feat_list.end();
         itPair++, count++)
    {
        // cout << "(" << itPair->other_x << "," << itPair->other_y << "," <<
        // itPair->this_z << ")" << "- (" << itPair->this_x << "," <<
        // itPair->this_y << "," << itPair->other_z << "): ";
        // cout << sqrt( square( itPair->other_x - itPair->this_x ) + square(
        // itPair->other_y - itPair->this_y ) + square( itPair->other_z -
        // itPair->this_z ) ) << endl;
        dist(count, 0) = sqrt(
            square(itPair->other_x - itPair->this_x) +
            square(itPair->other_y - itPair->this_y) +
            square(itPair->other_z - itPair->this_z));
    }

    dist.meanAndStdAll(v_mean, v_std);

    cout << endl
         << "*****************************************************" << endl;
    cout << "Mean: " << v_mean << " - STD: " << v_std << endl;
    cout << endl
         << "*****************************************************" << endl;

    // Filter out bad points
    unsigned int idx = 0;
    // for( int idx = (int)feat_list.size()-1; idx >= 0; idx-- )
    for (itPair = feat_list.begin(); itPair != feat_list.end(); idx++)
    {
        // if( dist( idx, 0 ) > 1.2 )
        if (fabs(dist(idx, 0) - v_mean) > v_std * numberOfSigmas)
        {
            cout << "Outlier deleted: " << dist(idx, 0) << " vs "
                 << v_std * numberOfSigmas << endl;
            itPair = feat_list.erase(itPair);
        }
        else
            itPair++;
    }

    MRPT_END
}  // end filterBadCorrsByDistance