main_vot.cpp

   1 #include <stdlib.h>
   2 #include <getopt.h>
   3 #include <libgen.h>
   4 #include <unistd.h>
   5
   6 #include "kcf.h"
   7 #include "vot.hpp"
   8
   9 double calcAccuracy(std::string line, cv::Rect bb_rect, cv::Rect &groundtruth_rect)
  10 {
  11     std::vector<float> numbers;
  12     std::istringstream s(line);
  13     float x;
  14     char ch;
  15
  16     while (s >> x) {
  17         numbers.push_back(x);
  18         s >> ch;
  19     }
  20     double x1 = std::min(numbers[0], std::min(numbers[2], std::min(numbers[4], numbers[6])));
  21     double x2 = std::max(numbers[0], std::max(numbers[2], std::max(numbers[4], numbers[6])));
  22     double y1 = std::min(numbers[1], std::min(numbers[3], std::min(numbers[5], numbers[7])));
  23     double y2 = std::max(numbers[1], std::max(numbers[3], std::max(numbers[5], numbers[7])));
  24
  25     groundtruth_rect = cv::Rect(x1, y1, x2 - x1, y2 - y1);
  26
  27     double rects_intersection = (groundtruth_rect & bb_rect).area();
  28     double rects_union = (groundtruth_rect | bb_rect).area();
  29     double accuracy = rects_intersection / rects_union;
  30
  31     return accuracy;
  32 }
  33
  34 int main(int argc, char *argv[])
  35 {
  36     // load region, images and prepare for output
  37     std::string region, images, output;
  38     int visualize_delay = -1, fit_size_x = -1, fit_size_y = -1;
  39     KCF_Tracker tracker;
  40
  41     while (1) {
  42         int option_index = 0;
  43         static struct option long_options[] = {{"debug", no_argument, 0, 'd'},
  44                                                {"visualDebug", no_argument, 0, 'p'},
  45                                                {"help", no_argument, 0, 'h'},
  46                                                {"output", required_argument, 0, 'o'},
  47                                                {"visualize", optional_argument, 0, 'v'},
  48                                                {"fit", optional_argument, 0, 'f'},
  49                                                {0, 0, 0, 0}};
  50
  51         int c = getopt_long(argc, argv, "dphv::f::o:", long_options, &option_index);
  52         if (c == -1) break;
  53
  54         switch (c) {
  55         case 'd':
  56             tracker.m_debug = true;
  57             break;
  58         case 'p':
  59             tracker.m_visual_debug = true;
  60             visualize_delay = 500;
  61             break;
  62         case 'h':
  63             std::cerr
  64                 << "Usage: \n"
  65                 << argv[0] << " [options]\n"
  66                 << argv[0] << " [options] <directory>\n"
  67                 << argv[0]
  68                 << " [options] <path/to/region.txt or groundtruth.txt> <path/to/images.txt> [path/to/output.txt]\n"
  69                 << "Options:\n"
  70                 << " --visualize | -v[delay_ms]\n"
  71                 << " --output    | -o <output.txt>\n"
  72                 << " --debug     | -d\n"
  73                 << " --visualDebug | -p\n"
  74                 << " --fit       | -f[WxH]\n";
  75             exit(0);
  76             break;
  77         case 'o':
  78             output = optarg;
  79             break;
  80         case 'v':
  81             visualize_delay = optarg ? atol(optarg) : 1;
  82             break;
  83         case 'f':
  84             std::string sizes = optarg ? optarg : "128x128";
  85             std::string delimiter = "x";
  86             size_t pos = sizes.find(delimiter);
  87             std::string first_argument = sizes.substr(0, pos);
  88             sizes.erase(0, pos + delimiter.length());
  89
  90             fit_size_x = stol(first_argument);
  91             fit_size_y = stol(sizes);
  92             break;
  93         }
  94     }
  95
  96     switch (argc - optind) {
  97     case 1:
  98         if (chdir(argv[optind]) == -1) {
  99             perror(argv[optind]);
 100             exit(1);
 101         }
 102         // Fall through
 103     case 0:
 104         region = access("groundtruth.txt", F_OK) == 0 ? "groundtruth.txt" : "region.txt";
 105         images = "images.txt";
 106         if (output.empty()) output = "output.txt";
 107         break;
 108     case 2:
 109         // Fall through
 110     case 3:
 111         region = std::string(argv[optind + 0]);
 112         images = std::string(argv[optind + 1]);
 113         if (output.empty()) {
 114             if ((argc - optind) == 3)
 115                 output = std::string(argv[optind + 2]);
 116             else
 117                 output = std::string(dirname(argv[optind + 0])) + "/output.txt";
 118         }
 119         break;
 120     default:
 121         std::cerr << "Too many arguments\n";
 122         return 1;
 123     }
 124     VOT vot_io(region, images, output);
 125
 126     // if groundtruth.txt is used use intersection over union (IOU) to calculate tracker accuracy
 127     std::ifstream groundtruth_stream;
 128     if (region.compare("groundtruth.txt") == 0) {
 129         groundtruth_stream.open(region.c_str());
 130         std::string line;
 131         std::getline(groundtruth_stream, line);
 132     }
 133
 134     cv::Mat image;
 135
 136     // img = firts frame, initPos = initial position in the first frame
 137     cv::Rect init_rect = vot_io.getInitRectangle();
 138     vot_io.outputBoundingBox(init_rect);
 139     vot_io.getNextImage(image);
 140
 141     tracker.init(image, init_rect, fit_size_x, fit_size_y);
 142
 143     BBox_c bb;
 144     cv::Rect bb_rect;
 145     double avg_time = 0., sum_accuracy = 0.;
 146     int frames = 0;
 147     while (vot_io.getNextImage(image) == 1) {
 148         double time_profile_counter = cv::getCPUTickCount();
 149         tracker.track(image);
 150         time_profile_counter = cv::getCPUTickCount() - time_profile_counter;
 151         std::cout << "  -> speed : " << time_profile_counter / ((double)cvGetTickFrequency() * 1000) << "ms. per frame";
 152         avg_time += time_profile_counter / ((double)cvGetTickFrequency() * 1000);
 153         frames++;
 154
 155         bb = tracker.getBBox();
 156         bb_rect = cv::Rect(bb.cx - bb.w / 2., bb.cy - bb.h / 2., bb.w, bb.h);
 157         vot_io.outputBoundingBox(bb_rect);
 158
 159         if (groundtruth_stream.is_open()) {
 160             std::string line;
 161             std::getline(groundtruth_stream, line);
 162
 163             cv::Rect groundtruthRect;
 164             double accuracy = calcAccuracy(line, bb_rect, groundtruthRect);
 165             if (visualize_delay >= 0) cv::rectangle(image, groundtruthRect, CV_RGB(255, 0, 0), 1);
 166             std::cout << ", accuracy: " << accuracy;
 167             sum_accuracy += accuracy;
 168         }
 169
 170         std::cout << std::endl;
 171
 172         if (visualize_delay >= 0) {
 173             cv::Point pt(bb.cx, bb.cy);
 174             cv::Size size(bb.w, bb.h);
 175             cv::RotatedRect rotatedRectangle(pt, size, bb.a);
 176
 177             cv::Point2f vertices[4];
 178             rotatedRectangle.points(vertices);
 179
 180             for (int i = 0; i < 4; i++)
 181                 cv::line(image, vertices[i], vertices[(i + 1) % 4], cv::Scalar(0, 255, 0), 2);
 182             //             cv::rectangle(image, cv::Rect(bb.cx - bb.w/2., bb.cy - bb.h/2., bb.w, bb.h), CV_RGB(0,255,0),
 183             //             2);
 184             std::string angle = std::to_string(bb.a);
 185             angle.erase(angle.find_last_not_of('0') + 1, std::string::npos);
 186             angle.erase(angle.find_last_not_of('.') + 1, std::string::npos);
 187             cv::putText(image, "Frame: " + std::to_string(frames) + " " + angle + " angle",
 188                         cv::Point(0, image.rows - 1), cv::FONT_HERSHEY_SIMPLEX, 0.7, cv::Scalar(0, 255, 0), 2);
 189             cv::imshow("output", image);
 190             int ret = cv::waitKey(visualize_delay);
 191             if (visualize_delay > 0 && ret != -1 && ret != 255) break;
 192         }
 193
 194         //        std::stringstream s;
 195         //        std::string ss;
 196         //        int countTmp = frames;
 197         //        s << "imgs" << "/img" << (countTmp/10000);
 198         //        countTmp = countTmp%10000;
 199         //        s << (countTmp/1000);
 200         //        countTmp = countTmp%1000;
 201         //        s << (countTmp/100);
 202         //        countTmp = countTmp%100;
 203         //        s << (countTmp/10);
 204         //        countTmp = countTmp%10;
 205         //        s << (countTmp);
 206         //        s << ".jpg";
 207         //        s >> ss;
 208         //        //set image output parameters
 209         //        std::vector<int> compression_params;
 210         //        compression_params.push_back(CV_IMWRITE_JPEG_QUALITY);
 211         //        compression_params.push_back(90);
 212         //        cv::imwrite(ss.c_str(), image, compression_params);
 213     }
 214
 215     std::cout << "Average processing speed " << avg_time / frames << "ms. (" << 1. / (avg_time / frames) * 1000
 216               << " fps)" << std::endl;
 217     if (groundtruth_stream.is_open()) {
 218         std::cout << "Average accuracy: " << sum_accuracy / frames << std::endl;
 219         groundtruth_stream.close();
 220     }
 221
 222     return EXIT_SUCCESS;
 223 }