perfGenericTracker.cpp

/****************************************************************************
 *
 * ViSP, open source Visual Servoing Platform software.
 * Copyright (C) 2005 - 2023 by Inria. All rights reserved.
 *
 * This software is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 * See the file LICENSE.txt at the root directory of this source
 * distribution for additional information about the GNU GPL.
 *
 * For using ViSP with software that can not be combined with the GNU
 * GPL, please contact Inria about acquiring a ViSP Professional
 * Edition License.
 *
 * See https://visp.inria.fr for more information.
 *
 * This software was developed at:
 * Inria Rennes - Bretagne Atlantique
 * Campus Universitaire de Beaulieu
 * 35042 Rennes Cedex
 * France
 *
 * If you have questions regarding the use of this file, please contact
 * Inria at visp@inria.fr
 *
 * This file is provided AS IS with NO WARRANTY OF ANY KIND, INCLUDING THE
 * WARRANTY OF DESIGN, MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE.
 *
 * Description:
 * Benchmark generic tracker.
 *
*****************************************************************************/
 
#include <visp3/core/vpConfig.h>
 
#if defined(VISP_HAVE_CATCH2)
#define CATCH_CONFIG_ENABLE_BENCHMARKING
#define CATCH_CONFIG_RUNNER
#include <catch.hpp>
 
#include <visp3/core/vpIoTools.h>
#include <visp3/io/vpImageIo.h>
#include <visp3/mbt/vpMbGenericTracker.h>
 
// #define DEBUG_DISPLAY // uncomment to check that the tracking is correct
#ifdef DEBUG_DISPLAY
#include <visp3/gui/vpDisplayX.h>
#endif
 
namespace
{
bool runBenchmark = false;
 
template <typename Type>
bool read_data(const std::string &input_directory, int cpt, const vpCameraParameters &cam_depth, vpImage<Type> &I,
               vpImage<uint16_t> &I_depth, std::vector<vpColVector> &pointcloud, vpHomogeneousMatrix &cMo)
{
  static_assert(std::is_same<Type, unsigned char>::value || std::is_same<Type, vpRGBa>::value,
                "Template function supports only unsigned char and vpRGBa images!");
#if VISP_HAVE_DATASET_VERSION >= 0x030600
  std::string ext("png");
#else
  std::string ext("pgm");
#endif
  char buffer[FILENAME_MAX];
  snprintf(buffer, FILENAME_MAX, std::string(input_directory + "/Images/Image_%04d." + ext).c_str(), cpt);
  std::string image_filename = buffer;
 
  snprintf(buffer, FILENAME_MAX, std::string(input_directory + "/Depth/Depth_%04d.bin").c_str(), cpt);
  std::string depth_filename = buffer;
 
  snprintf(buffer, FILENAME_MAX, std::string(input_directory + "/CameraPose/Camera_%03d.txt").c_str(), cpt);
  std::string pose_filename = buffer;
 
  if (!vpIoTools::checkFilename(image_filename) || !vpIoTools::checkFilename(depth_filename) ||
      !vpIoTools::checkFilename(pose_filename))
    return false;
 
  vpImageIo::read(I, image_filename);
 
  unsigned int depth_width = 0, depth_height = 0;
  std::ifstream file_depth(depth_filename.c_str(), std::ios::in | std::ios::binary);
  if (!file_depth.is_open())
    return false;
 
  vpIoTools::readBinaryValueLE(file_depth, depth_height);
  vpIoTools::readBinaryValueLE(file_depth, depth_width);
  I_depth.resize(depth_height, depth_width);
  pointcloud.resize(depth_height * depth_width);
 
  const float depth_scale = 0.000030518f;
  for (unsigned int i = 0; i < I_depth.getHeight(); i++) {
    for (unsigned int j = 0; j < I_depth.getWidth(); j++) {
      vpIoTools::readBinaryValueLE(file_depth, I_depth[i][j]);
      double x = 0.0, y = 0.0, Z = I_depth[i][j] * depth_scale;
      vpPixelMeterConversion::convertPoint(cam_depth, j, i, x, y);
      vpColVector pt3d(4, 1.0);
      pt3d[0] = x * Z;
      pt3d[1] = y * Z;
      pt3d[2] = Z;
      pointcloud[i * I_depth.getWidth() + j] = pt3d;
    }
  }
 
  std::ifstream file_pose(pose_filename.c_str());
  if (!file_pose.is_open()) {
    return false;
  }
 
  for (unsigned int i = 0; i < 4; i++) {
    for (unsigned int j = 0; j < 4; j++) {
      file_pose >> cMo[i][j];
    }
  }
 
  return true;
}
} // anonymous namespace
 
TEST_CASE("Benchmark generic tracker", "[benchmark]")
{
  if (runBenchmark) {
    std::vector<int> tracker_type(2);
    tracker_type[0] = vpMbGenericTracker::EDGE_TRACKER;
    tracker_type[1] = vpMbGenericTracker::DEPTH_DENSE_TRACKER;
    vpMbGenericTracker tracker(tracker_type);
 
    const std::string input_directory =
      vpIoTools::createFilePath(vpIoTools::getViSPImagesDataPath(), "mbt-depth/Castle-simu");
 
    const bool verbose = false;
#if defined(VISP_HAVE_PUGIXML)
    const std::string configFileCam1 = input_directory + std::string("/Config/chateau.xml");
    const std::string configFileCam2 = input_directory + std::string("/Config/chateau_depth.xml");
    REQUIRE(vpIoTools::checkFilename(configFileCam1));
    REQUIRE(vpIoTools::checkFilename(configFileCam2));
    tracker.loadConfigFile(configFileCam1, configFileCam2, verbose);
#else
    {
      vpCameraParameters cam_color, cam_depth;
      cam_color.initPersProjWithoutDistortion(700.0, 700.0, 320.0, 240.0);
      cam_depth.initPersProjWithoutDistortion(700.0, 700.0, 320.0, 240.0);
      tracker.setCameraParameters(cam_color, cam_depth);
    }
 
    // Edge
    vpMe me;
    me.setMaskSize(5);
    me.setMaskNumber(180);
    me.setRange(8);
    me.setLikelihoodThresholdType(vpMe::NORMALIZED_THRESHOLD);
    me.setThreshold(5);
    me.setMu1(0.5);
    me.setMu2(0.5);
    me.setSampleStep(5);
    tracker.setMovingEdge(me);
 
    // Klt
#if defined(VISP_HAVE_MODULE_KLT) && defined(VISP_HAVE_OPENCV) && defined(HAVE_OPENCV_IMGPROC) && defined(HAVE_OPENCV_VIDEO)
    vpKltOpencv klt;
    tracker.setKltMaskBorder(5);
    klt.setMaxFeatures(10000);
    klt.setWindowSize(5);
    klt.setQuality(0.01);
    klt.setMinDistance(5);
    klt.setHarrisFreeParameter(0.02);
    klt.setBlockSize(3);
    klt.setPyramidLevels(3);
 
    tracker.setKltOpencv(klt);
#endif
 
    // Depth
    tracker.setDepthNormalFeatureEstimationMethod(vpMbtFaceDepthNormal::ROBUST_FEATURE_ESTIMATION);
    tracker.setDepthNormalPclPlaneEstimationMethod(2);
    tracker.setDepthNormalPclPlaneEstimationRansacMaxIter(200);
    tracker.setDepthNormalPclPlaneEstimationRansacThreshold(0.001);
    tracker.setDepthNormalSamplingStep(2, 2);
 
    tracker.setDepthDenseSamplingStep(4, 4);
 
    tracker.setAngleAppear(vpMath::rad(85.0));
    tracker.setAngleDisappear(vpMath::rad(89.0));
    tracker.setNearClippingDistance(0.01);
    tracker.setFarClippingDistance(2.0);
    tracker.setClipping(tracker.getClipping() | vpMbtPolygon::FOV_CLIPPING);
#endif
 
    REQUIRE(vpIoTools::checkFilename(input_directory + "/Models/chateau.cao"));
    tracker.loadModel(input_directory + "/Models/chateau.cao", input_directory + "/Models/chateau.cao", verbose);
 
    vpHomogeneousMatrix T;
    T[0][0] = -1;
    T[0][3] = -0.2;
    T[1][1] = 0;
    T[1][2] = 1;
    T[1][3] = 0.12;
    T[2][1] = 1;
    T[2][2] = 0;
    T[2][3] = -0.15;
    tracker.loadModel(input_directory + "/Models/cube.cao", verbose, T);
 
    vpImage<unsigned char> I;
    vpImage<uint16_t> I_depth_raw;
    vpHomogeneousMatrix cMo_truth;
    std::vector<vpColVector> pointcloud;
 
    vpCameraParameters cam_color, cam_depth;
    tracker.getCameraParameters(cam_color, cam_depth);
 
    vpHomogeneousMatrix depth_M_color;
    depth_M_color[0][3] = -0.05;
    tracker.setCameraTransformationMatrix("Camera2", depth_M_color);
 
    // load all the data in memory to not take into account I/O from disk
    std::vector<vpImage<unsigned char> > images;
    std::vector<vpImage<uint16_t> > depth_raws;
    std::vector<std::vector<vpColVector> > pointclouds;
    std::vector<vpHomogeneousMatrix> cMo_truth_all;
    // forward
    for (int i = 1; i <= 40; i++) {
      if (read_data(input_directory, i, cam_depth, I, I_depth_raw, pointcloud, cMo_truth)) {
        images.push_back(I);
        depth_raws.push_back(I_depth_raw);
        pointclouds.push_back(pointcloud);
        cMo_truth_all.push_back(cMo_truth);
      }
    }
    // backward
    for (int i = 40; i >= 1; i--) {
      if (read_data(input_directory, i, cam_depth, I, I_depth_raw, pointcloud, cMo_truth)) {
        images.push_back(I);
        depth_raws.push_back(I_depth_raw);
        pointclouds.push_back(pointcloud);
        cMo_truth_all.push_back(cMo_truth);
      }
    }
 
    // Stereo MBT
    {
      std::vector<std::map<std::string, int> > mapOfTrackerTypes;
      mapOfTrackerTypes.push_back(
          { {"Camera1", vpMbGenericTracker::EDGE_TRACKER}, {"Camera2", vpMbGenericTracker::DEPTH_DENSE_TRACKER} });
      mapOfTrackerTypes.push_back(
          { {"Camera1", vpMbGenericTracker::EDGE_TRACKER}, {"Camera2", vpMbGenericTracker::DEPTH_DENSE_TRACKER} });
#if defined(VISP_HAVE_OPENCV) && defined(HAVE_OPENCV_IMGPROC) && defined(HAVE_OPENCV_VIDEO)
      mapOfTrackerTypes.push_back(
          { {"Camera1", vpMbGenericTracker::KLT_TRACKER}, {"Camera2", vpMbGenericTracker::DEPTH_DENSE_TRACKER} });
      mapOfTrackerTypes.push_back({ {"Camera1", vpMbGenericTracker::EDGE_TRACKER | vpMbGenericTracker::KLT_TRACKER},
                                   {"Camera2", vpMbGenericTracker::DEPTH_DENSE_TRACKER} });
      mapOfTrackerTypes.push_back({ {"Camera1", vpMbGenericTracker::EDGE_TRACKER | vpMbGenericTracker::KLT_TRACKER},
                                   {"Camera2", vpMbGenericTracker::DEPTH_DENSE_TRACKER} });
#endif
 
      std::vector<std::string> benchmarkNames = {
        "Edge MBT",
        "Edge + Depth dense MBT",
#if defined(VISP_HAVE_OPENCV)
        "KLT MBT",
        "KLT + depth dense MBT",
        "Edge + KLT + depth dense MBT"
#endif
      };
 
      std::vector<bool> monoculars = {
        true,
        false,
#if defined(VISP_HAVE_OPENCV)
        true,
        false,
        false
#endif
      };
 
      for (size_t idx = 0; idx < mapOfTrackerTypes.size(); idx++) {
        tracker.resetTracker();
        tracker.setTrackerType(mapOfTrackerTypes[idx]);
 
        const bool verbose = false;
#if defined(VISP_HAVE_PUGIXML)
        tracker.loadConfigFile(configFileCam1, configFileCam2, verbose);
#else
        {
          vpCameraParameters cam_color, cam_depth;
          cam_color.initPersProjWithoutDistortion(700.0, 700.0, 320.0, 240.0);
          cam_depth.initPersProjWithoutDistortion(700.0, 700.0, 320.0, 240.0);
          tracker.setCameraParameters(cam_color, cam_depth);
        }
 
        // Edge
        vpMe me;
        me.setMaskSize(5);
        me.setMaskNumber(180);
        me.setRange(8);
        me.setLikelihoodThresholdType(vpMe::NORMALIZED_THRESHOLD);
        me.setThreshold(5);
        me.setMu1(0.5);
        me.setMu2(0.5);
        me.setSampleStep(5);
        tracker.setMovingEdge(me);
 
        // Klt
#if defined(VISP_HAVE_MODULE_KLT) && defined(VISP_HAVE_OPENCV) && defined(HAVE_OPENCV_IMGPROC) && defined(HAVE_OPENCV_VIDEO)
        vpKltOpencv klt;
        tracker.setKltMaskBorder(5);
        klt.setMaxFeatures(10000);
        klt.setWindowSize(5);
        klt.setQuality(0.01);
        klt.setMinDistance(5);
        klt.setHarrisFreeParameter(0.02);
        klt.setBlockSize(3);
        klt.setPyramidLevels(3);
 
        tracker.setKltOpencv(klt);
#endif
 
        // Depth
        tracker.setDepthNormalFeatureEstimationMethod(vpMbtFaceDepthNormal::ROBUST_FEATURE_ESTIMATION);
        tracker.setDepthNormalPclPlaneEstimationMethod(2);
        tracker.setDepthNormalPclPlaneEstimationRansacMaxIter(200);
        tracker.setDepthNormalPclPlaneEstimationRansacThreshold(0.001);
        tracker.setDepthNormalSamplingStep(2, 2);
 
        tracker.setDepthDenseSamplingStep(4, 4);
 
        tracker.setAngleAppear(vpMath::rad(85.0));
        tracker.setAngleDisappear(vpMath::rad(89.0));
        tracker.setNearClippingDistance(0.01);
        tracker.setFarClippingDistance(2.0);
        tracker.setClipping(tracker.getClipping() | vpMbtPolygon::FOV_CLIPPING);
#endif
        tracker.loadModel(input_directory + "/Models/chateau.cao", input_directory + "/Models/chateau.cao", verbose);
        tracker.loadModel(input_directory + "/Models/cube.cao", verbose, T);
        tracker.initFromPose(images.front(), cMo_truth_all.front());
 
        std::map<std::string, unsigned int> mapOfWidths, mapOfHeights;
        mapOfWidths["Camera2"] = monoculars[idx] ? 0 : I_depth_raw.getWidth();
        mapOfHeights["Camera2"] = monoculars[idx] ? 0 : I_depth_raw.getHeight();
 
        vpHomogeneousMatrix cMo;
#ifndef DEBUG_DISPLAY
        BENCHMARK(benchmarkNames[idx].c_str())
#else
        vpImage<unsigned char> I_depth;
        vpImageConvert::createDepthHistogram(I_depth_raw, I_depth);
 
        vpDisplayX d_color(I, 0, 0, "Color image");
        vpDisplayX d_depth(I_depth, I.getWidth(), 0, "Depth image");
        tracker.setDisplayFeatures(true);
#endif
        {
          tracker.initFromPose(images.front(), cMo_truth_all.front());
 
          for (size_t i = 0; i < images.size(); i++) {
            const vpImage<unsigned char> &I_current = images[i];
            const std::vector<vpColVector> &pointcloud_current = pointclouds[i];
 
#ifdef DEBUG_DISPLAY
            vpImageConvert::createDepthHistogram(depth_raws[i], I_depth);
            I = I_current;
            vpDisplay::display(I);
            vpDisplay::display(I_depth);
#endif
 
            std::map<std::string, const vpImage<unsigned char> *> mapOfImages;
            mapOfImages["Camera1"] = &I_current;
 
            std::map<std::string, const std::vector<vpColVector> *> mapOfPointclouds;
            mapOfPointclouds["Camera2"] = &pointcloud_current;
 
            tracker.track(mapOfImages, mapOfPointclouds, mapOfWidths, mapOfHeights);
            cMo = tracker.getPose();
 
#ifdef DEBUG_DISPLAY
            tracker.display(I, I_depth, cMo, depth_M_color * cMo, cam_color, cam_depth, vpColor::red, 3);
            vpDisplay::displayFrame(I, cMo, cam_color, 0.05, vpColor::none, 3);
            vpDisplay::displayFrame(I_depth, depth_M_color * cMo, cam_depth, 0.05, vpColor::none, 3);
            vpDisplay::displayText(I, 20, 20, benchmarkNames[idx], vpColor::red);
            vpDisplay::displayText(
                I, 40, 20, std::string("Nb features: " + std::to_string(tracker.getError().getRows())), vpColor::red);
 
            vpDisplay::flush(I);
            vpDisplay::flush(I_depth);
            vpTime::wait(33);
#endif
          }
 
#ifndef DEBUG_DISPLAY
          return cMo;
        };
#else
      }
#endif
 
        vpPoseVector pose_est(cMo);
        vpPoseVector pose_truth(cMo_truth);
        vpColVector t_err(3), tu_err(3);
        for (unsigned int i = 0; i < 3; i++) {
          t_err[i] = pose_est[i] - pose_truth[i];
          tu_err[i] = pose_est[i + 3] - pose_truth[i + 3];
        }
 
        const double max_translation_error = 0.006;
        const double max_rotation_error = 0.03;
        CHECK(sqrt(t_err.sumSquare()) < max_translation_error);
        CHECK(sqrt(tu_err.sumSquare()) < max_rotation_error);
    }
  }
} // if (runBenchmark)
}
 
int main(int argc, char *argv[])
{
  Catch::Session session; // There must be exactly one instance
 
  // Build a new parser on top of Catch's
  using namespace Catch::clara;
  auto cli = session.cli()         // Get Catch's composite command line parser
    | Opt(runBenchmark)   // bind variable to a new option, with a hint string
    ["--benchmark"] // the option names it will respond to
    ("run benchmark comparing naive code with ViSP implementation"); // description string for the help output
 
// Now pass the new composite back to Catch so it uses that
  session.cli(cli);
 
  // Let Catch (using Clara) parse the command line
  session.applyCommandLine(argc, argv);
 
  int numFailed = session.run();
 
  // numFailed is clamped to 255 as some unices only use the lower 8 bits.
  // This clamping has already been applied, so just return it here
  // You can also do any post run clean-up here
  return numFailed;
}
#else
#include <iostream>
 
int main() { return EXIT_SUCCESS; }
#endif