CObservationVelodyneScan.h

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 |
   +------------------------------------------------------------------------+ */
#ifndef CObservationVelodyneScan_H
#define CObservationVelodyneScan_H

#include <mrpt/serialization/CSerializable.h>
#include <mrpt/obs/CObservation.h>
#include <mrpt/obs/CSinCosLookUpTableFor2DScans.h>
#include <mrpt/poses/CPose3D.h>
#include <mrpt/obs/VelodyneCalibration.h>
#include <vector>

namespace mrpt
{
namespace poses
{
class CPose3DInterpolator;
}
namespace obs
{
/** A `CObservation`-derived class for RAW DATA (and optionally, point cloud) of
 * scans from 3D Velodyne LIDAR scanners.
 * A scan comprises one or more "velodyne packets" (refer to Velodyne user
 * manual).
 *
 * <h2>Main data fields:</h2><hr>
 * - CObservationVelodyneScan::scan_packets with raw data packets.
 * - CObservationVelodyneScan::point_cloud normally empty after grabbing for
 * efficiency, can be generated calling \a
 * CObservationVelodyneScan::generatePointCloud()
 *
 * Axes convention for point cloud (x,y,z) coordinates:
 *
 *  <div align=center> <img src="velodyne_axes.jpg"> </div>
 *
 * If it can be assumed that the sensor moves SLOWLY through the environment
 * (i.e. its pose can be approximated to be the same since the beginning to the
 * end of one complete scan)
 * then this observation can be converted / loaded into the following other
 * classes:
 *  - Maps of points (these require first generating the pointcloud in this
 * observation object with
 * mrpt::obs::CObservationVelodyneScan::generatePointCloud() ):
 *    - mrpt::maps::CPointsMap::loadFromVelodyneScan() (available in all
 * derived classes)
 *    - and the generic method:mrpt::maps::CPointsMap::insertObservation()
 *  - mrpt::opengl::CPointCloud and mrpt::opengl::CPointCloudColoured is
 * supported by first converting
 *    this scan to a mrpt::maps::CPointsMap-derived class, then loading it into
 * the opengl object.
 *
 * Otherwise, the following API exists for accurate reconstruction of the
 * sensor path in SE(3) over time:
 *  - CObservationVelodyneScan::generatePointCloudAlongSE3Trajectory()
 *
 *  Note that this object has \b two timestamp fields:
 *  - The standard CObservation::timestamp field in the base class, which
 * should contain the accurate satellite-based UTC timestamp, and
 *  - the field CObservationVelodyneScan::originalReceivedTimestamp, with the
 * local computer-based timestamp based on the reception of the message in the
 * computer.
 *  Both timestamps correspond to the firing of the <b>first</b> laser in the
 * <b>first</b> CObservationVelodyneScan::scan_packets packet.
 *
 * \note New in MRPT 1.4.0
 * \sa CObservation, CPointsMap, CVelodyneScanner
 * \ingroup mrpt_obs_grp
 */
class CObservationVelodyneScan : public CObservation
{
    DEFINE_SERIALIZABLE(CObservationVelodyneScan)

   public:
    /** @name Raw scan fixed parameters
        @{ */
    static const int SIZE_BLOCK = 100;
    static const int RAW_SCAN_SIZE = 3;
    static const int SCANS_PER_BLOCK = 32;
    static const int BLOCK_DATA_SIZE = (SCANS_PER_BLOCK * RAW_SCAN_SIZE);

    static const float ROTATION_RESOLUTION;  // = 0.01f; /**< degrees */
    static const uint16_t ROTATION_MAX_UNITS =
        36000; /**< hundredths of degrees */

    static const float DISTANCE_MAX; /**< meters */
    static const float DISTANCE_RESOLUTION; /**< meters */
    static const float DISTANCE_MAX_UNITS;

    /** Blocks 0-31 */
    static const uint16_t UPPER_BANK = 0xeeff;
    /** Blocks 32-63 */
    static const uint16_t LOWER_BANK = 0xddff;

    static const int PACKET_SIZE = 1206;
    static const int POS_PACKET_SIZE = 512;
    static const int BLOCKS_PER_PACKET = 12;
    static const int PACKET_STATUS_SIZE = 4;
    static const int SCANS_PER_PACKET = (SCANS_PER_BLOCK * BLOCKS_PER_PACKET);

    static const uint8_t RETMODE_STRONGEST = 0x37;
    static const uint8_t RETMODE_LAST = 0x38;
    static const uint8_t RETMODE_DUAL = 0x39;
/** @} */

/** @name Scan data
    @{ */

#pragma pack(push, 1)
    struct laser_return_t
    {
        uint16_t distance;
        uint8_t intensity;
    };
    /** Raw Velodyne data block.
     *  Each block contains data from either the upper or lower laser
     *  bank.  The device returns three times as many upper bank blocks. */
    struct raw_block_t
    {
        uint16_t header;  ///< Block id: UPPER_BANK or LOWER_BANK
        uint16_t rotation;  ///< 0-35999, divide by 100 to get degrees
        laser_return_t laser_returns[SCANS_PER_BLOCK];
    };

    /** One unit of data from the scanner (the payload of one UDP DATA packet)
     */
    struct TVelodyneRawPacket
    {
        raw_block_t blocks[BLOCKS_PER_PACKET];
        /** us from top of hour */
        uint32_t gps_timestamp;
        /** 0x37: strongest, 0x38: last, 0x39: dual return */
        uint8_t laser_return_mode;
        /** 0x21: HDL-32E, 0x22: VLP-16 */
        uint8_t velodyne_model_ID;
    };

    /** Payload of one POSITION packet */
    struct TVelodynePositionPacket
    {
        char unused1[198];
        uint32_t gps_timestamp;
        uint32_t unused2;
        /** the full $GPRMC message, as received by Velodyne, terminated with
         * "\r\n\0" */
        char NMEA_GPRMC[72 + 234];
    };
#pragma pack(pop)

    /** The maximum range allowed by the device, in meters (e.g. 100m). Stored
     * here by the driver while capturing based on the sensor model. */
    double minRange{1.0}, maxRange{130.0};
    /** The 6D pose of the sensor on the robot/vehicle frame of reference */
    mrpt::poses::CPose3D sensorPose;
    /** The main content of this object: raw data packet from the LIDAR. \sa
     * point_cloud */
    std::vector<TVelodyneRawPacket> scan_packets;
    /** The calibration data for the LIDAR device. See
     * mrpt::hwdrivers::CVelodyneScanner and mrpt::obs::VelodyneCalibration for
     * details. */
    mrpt::obs::VelodyneCalibration calibration;
    /** The local computer-based timestamp based on the reception of the message
     * in the computer. \sa has_satellite_timestamp, CObservation::timestamp in
     * the base class, which should contain the accurate satellite-based UTC
     * timestamp.  */
    mrpt::system::TTimeStamp originalReceivedTimestamp{INVALID_TIMESTAMP};
    /** If true, CObservation::timestamp has been generated from accurate
     * satellite clock. Otherwise, no GPS data is available and timestamps are
     * based on the local computer clock. */
    bool has_satellite_timestamp{false};

    mrpt::system::TTimeStamp getOriginalReceivedTimeStamp()
        const override;  // See base class docs

    /** See \a point_cloud and \a scan_packets */
    struct TPointCloud
    {
        std::vector<float> x, y, z;
        /** Color [0,255] */
        std::vector<uint8_t> intensity;
        /** Timestamp for each point (if `generatePerPointTimestamp`=true in
         * `TGeneratePointCloudParameters`), or empty vector if not populated
         * (default). */
        std::vector<mrpt::system::TTimeStamp> timestamp;
        /** Original azimuth of each point (if `generatePerPointAzimuth`=true,
         * empty otherwise ) */
        std::vector<float> azimuth;

        inline size_t size() const { return x.size(); }
        /** Sets all vectors to zero length */
        void clear();
        /** Like clear(), but also enforcing freeing memory */
        void clear_deep();
    };

    /** Optionally, raw data can be converted into a 3D point cloud (local
     * coordinates wrt the sensor, not the vehicle)
     * with intensity (graylevel) information. See axes convention in
     * mrpt::obs::CObservationVelodyneScan \sa generatePointCloud()
     */
    TPointCloud point_cloud;
    /** @} */

    /** @name Related to conversion to 3D point cloud
     * @{ */
    struct TGeneratePointCloudParameters
    {
        TGeneratePointCloudParameters() {}
        /** Minimum azimuth, in degrees (Default=0). Points will be generated
         * only the the area of interest [minAzimuth, maxAzimuth] */
        double minAzimuth_deg{.0};
        /** Minimum azimuth, in degrees (Default=360). Points will be generated
         * only the the area of interest [minAzimuth, maxAzimuth] */
        double maxAzimuth_deg{360.};
        /** Minimum (default=1.0f) and maximum (default: Infinity)
         * distances/ranges for a point to be considered. Points must pass this
         * (application specific) condition and also the minRange/maxRange
         * values in CObservationVelodyneScan (sensor-specific). */
        float minDistance{1.0f}, maxDistance{std::numeric_limits<float>::max()};
        /** The limits of the 3D box (default=infinity) in sensor (not vehicle)
         * local coordinates for the ROI filter \sa filterByROI */
        float ROI_x_min{-std::numeric_limits<float>::max()},
            ROI_x_max{std::numeric_limits<float>::max()},
            ROI_y_min{-std::numeric_limits<float>::max()},
            ROI_y_max{std::numeric_limits<float>::max()},
            ROI_z_min{-std::numeric_limits<float>::max()},
            ROI_z_max{std::numeric_limits<float>::max()};
        /** The limits of the 3D box (default=0) in sensor (not vehicle) local
         * coordinates for the nROI filter \sa filterBynROI */
        float nROI_x_min{0}, nROI_x_max{0}, nROI_y_min{0}, nROI_y_max{0},
            nROI_z_min{0}, nROI_z_max{0};
        /** (Default:2.0 meters) Minimum distance between a point and its two
         * neighbors to be considered an invalid point. */
        float isolatedPointsFilterDistance{2.0f};

        /** Enable ROI filter (Default:false): add points inside a given 3D box
         */
        bool filterByROI{false};
        /** Enable nROI filter (Default:false): do NOT add points inside a given
         * 3D box */
        bool filterBynROI{false};
        /** (Default:false) Simple filter to remove spurious returns (e.g. Sun
         * reflected on large water extensions) */
        bool filterOutIsolatedPoints{false};
        /** (Default:true) In VLP16 dual mode, keep both or just one of the
         * returns. */
        bool dualKeepStrongest{true}, dualKeepLast{true};
        /** (Default:false) If `true`, populate the vector timestamp */
        bool generatePerPointTimestamp{false};
        /** (Default:false) If `true`, populate the vector azimuth */
        bool generatePerPointAzimuth{false};
    };

    /** Generates the point cloud into the point cloud data fields in \a
     * CObservationVelodyneScan::point_cloud
     * where it is stored in local coordinates wrt the sensor (neither the
     * vehicle nor the world).
     * So, this method does not take into account the possible motion of the
     * sensor through the world as it collects LIDAR scans.
     * For high dynamics, see the more costly API
     * generatePointCloudAlongSE3Trajectory()
     * \note Points with ranges out of [minRange,maxRange] are discarded; as
     * well, other filters are available in \a params.
     * \sa generatePointCloudAlongSE3Trajectory(),
     * TGeneratePointCloudParameters
     */
    void generatePointCloud(
        const TGeneratePointCloudParameters& params =
            TGeneratePointCloudParameters());

    /** Results for generatePointCloudAlongSE3Trajectory() */
    struct TGeneratePointCloudSE3Results
    {
        /** Number of points in the observation */
        size_t num_points{0};
        /** Number of points for which a valid interpolated SE(3) pose could be
         * determined */
        size_t num_correctly_inserted_points{0};
    };

    /** An alternative to generatePointCloud() for cases where the motion of the
     * sensor as it grabs one scan (360 deg horiz FOV) cannot be ignored.
     * \param[in] vehicle_path Timestamped sequence of known poses for the
     * VEHICLE. Recall that the sensor has a relative pose wrt the vehicle
     * according to CObservationVelodyneScan::getSensorPose() &
     * CObservationVelodyneScan::setSensorPose()
     * \param[out] out_points The generated points, in the same coordinates
     * frame than \a vehicle_path. Points are APPENDED to the list, so prevous
     * contents are kept.
     * \param[out] results_stats Stats
     * \param[in] params Filtering and other parameters
     * \sa generatePointCloud(), TGeneratePointCloudParameters
     */
    void generatePointCloudAlongSE3Trajectory(
        const mrpt::poses::CPose3DInterpolator& vehicle_path,
        std::vector<mrpt::math::TPointXYZIu8>& out_points,
        TGeneratePointCloudSE3Results& results_stats,
        const TGeneratePointCloudParameters& params =
            TGeneratePointCloudParameters());

    /** @} */

    void getSensorPose(mrpt::poses::CPose3D& out_sensorPose) const override
    {
        out_sensorPose = sensorPose;
    }  // See base class docs
    void setSensorPose(const mrpt::poses::CPose3D& newSensorPose) override
    {
        sensorPose = newSensorPose;
    }  // See base class docs
    void getDescriptionAsText(
        std::ostream& o) const override;  // See base class docs

};  // End of class def.

}  // namespace obs

namespace typemeta
{  // Specialization must occur in the same namespace
MRPT_DECLARE_TTYPENAME_PTR_NAMESPACE(CObservationVelodyneScan, ::mrpt::obs)
}

}  // namespace mrpt

#endif