main_vot.cpp

   1 #include <stdlib.h>
   2 #include <getopt.h>
   3 #include <libgen.h>
   4 #include <unistd.h>
   5 #include <iomanip>
   6
   7 #include "kcf.h"
   8 #include "vot.hpp"
   9
  10 double calcAccuracy(std::string line, cv::Rect bb_rect, cv::Rect &groundtruth_rect)
  11 {
  12     std::vector<float> numbers;
  13     std::istringstream s(line);
  14     float x;
  15     char ch;
  16
  17     while (s >> x) {
  18         numbers.push_back(x);
  19         s >> ch;
  20     }
  21     double x1 = std::min(numbers[0], std::min(numbers[2], std::min(numbers[4], numbers[6])));
  22     double x2 = std::max(numbers[0], std::max(numbers[2], std::max(numbers[4], numbers[6])));
  23     double y1 = std::min(numbers[1], std::min(numbers[3], std::min(numbers[5], numbers[7])));
  24     double y2 = std::max(numbers[1], std::max(numbers[3], std::max(numbers[5], numbers[7])));
  25
  26     groundtruth_rect = cv::Rect(x1, y1, x2 - x1, y2 - y1);
  27
  28     double rects_intersection = (groundtruth_rect & bb_rect).area();
  29     double rects_union = (groundtruth_rect | bb_rect).area();
  30     double accuracy = rects_intersection / rects_union;
  31
  32     return accuracy;
  33 }
  34
  35 int main(int argc, char *argv[])
  36 {
  37     // load region, images and prepare for output
  38     std::string region, images, output;
  39     int visualize_delay = -1, fit_size_x = -1, fit_size_y = -1;
  40     KCF_Tracker tracker;
  41
  42     while (1) {
  43         int option_index = 0;
  44         static struct option long_options[] = {{"debug", no_argument, 0, 'd'},
  45                                                {"visualDebug", no_argument, 0, 'p'},
  46                                                {"help", no_argument, 0, 'h'},
  47                                                {"output", required_argument, 0, 'o'},
  48                                                {"visualize", optional_argument, 0, 'v'},
  49                                                {"fit", optional_argument, 0, 'f'},
  50                                                {0, 0, 0, 0}};
  51
  52         int c = getopt_long(argc, argv, "dphv::f::o:", long_options, &option_index);
  53         if (c == -1) break;
  54
  55         switch (c) {
  56         case 'd':
  57             tracker.m_debug = true;
  58             break;
  59         case 'p':
  60             tracker.m_visual_debug = true;
  61             visualize_delay = 500;
  62             break;
  63         case 'h':
  64             std::cerr
  65                 << "Usage: \n"
  66                 << argv[0] << " [options]\n"
  67                 << argv[0] << " [options] <directory>\n"
  68                 << argv[0]
  69                 << " [options] <path/to/region.txt or groundtruth.txt> <path/to/images.txt> [path/to/output.txt]\n"
  70                 << "Options:\n"
  71                 << " --visualize | -v[delay_ms]\n"
  72                 << " --output    | -o <output.txt>\n"
  73                 << " --debug     | -d\n"
  74                 << " --visualDebug | -p\n"
  75                 << " --fit       | -f[WxH]\n";
  76             exit(0);
  77             break;
  78         case 'o':
  79             output = optarg;
  80             break;
  81         case 'v':
  82             visualize_delay = optarg ? atol(optarg) : 1;
  83             break;
  84         case 'f':
  85             std::string sizes = optarg ? optarg : "128x128";
  86             std::string delimiter = "x";
  87             size_t pos = sizes.find(delimiter);
  88             std::string first_argument = sizes.substr(0, pos);
  89             sizes.erase(0, pos + delimiter.length());
  90
  91             fit_size_x = stol(first_argument);
  92             fit_size_y = stol(sizes);
  93             break;
  94         }
  95     }
  96
  97     switch (argc - optind) {
  98     case 1:
  99         if (chdir(argv[optind]) == -1) {
 100             perror(argv[optind]);
 101             exit(1);
 102         }
 103         // Fall through
 104     case 0:
 105         region = access("groundtruth.txt", F_OK) == 0 ? "groundtruth.txt" : "region.txt";
 106         images = "images.txt";
 107         if (output.empty()) output = "output.txt";
 108         break;
 109     case 2:
 110         // Fall through
 111     case 3:
 112         region = std::string(argv[optind + 0]);
 113         images = std::string(argv[optind + 1]);
 114         if (output.empty()) {
 115             if ((argc - optind) == 3)
 116                 output = std::string(argv[optind + 2]);
 117             else
 118                 output = std::string(dirname(argv[optind + 0])) + "/output.txt";
 119         }
 120         break;
 121     default:
 122         std::cerr << "Too many arguments\n";
 123         return 1;
 124     }
 125     VOT vot_io(region, images, output);
 126
 127     // if groundtruth.txt is used use intersection over union (IOU) to calculate tracker accuracy
 128     std::ifstream groundtruth_stream;
 129     if (region.compare("groundtruth.txt") == 0) {
 130         groundtruth_stream.open(region.c_str());
 131         std::string line;
 132         std::getline(groundtruth_stream, line);
 133     }
 134
 135     cv::Mat image;
 136
 137     // img = firts frame, initPos = initial position in the first frame
 138     cv::Rect init_rect = vot_io.getInitRectangle();
 139     vot_io.outputBoundingBox(init_rect);
 140     vot_io.getNextImage(image);
 141
 142     tracker.init(image, init_rect, fit_size_x, fit_size_y);
 143
 144     BBox_c bb;
 145     cv::Rect bb_rect;
 146     double avg_time = 0., sum_accuracy = 0.;
 147     int frames = 0;
 148
 149     std::cout << std::fixed << std::setprecision(2);
 150
 151     while (vot_io.getNextImage(image) == 1){
 152         double time_profile_counter = cv::getCPUTickCount();
 153         tracker.track(image);
 154         time_profile_counter = cv::getCPUTickCount() - time_profile_counter;
 155          std::cout << "  -> speed : " <<  time_profile_counter/((double)cvGetTickFrequency()*1000) << "ms per frame, "
 156                       "response : " << tracker.getFilterResponse();
 157         avg_time += time_profile_counter/((double)cvGetTickFrequency()*1000);
 158         frames++;
 159
 160         bb = tracker.getBBox();
 161         bb_rect = cv::Rect(bb.cx - bb.w / 2., bb.cy - bb.h / 2., bb.w, bb.h);
 162         vot_io.outputBoundingBox(bb_rect);
 163
 164         if (groundtruth_stream.is_open()) {
 165             std::string line;
 166             std::getline(groundtruth_stream, line);
 167
 168             cv::Rect groundtruthRect;
 169             double accuracy = calcAccuracy(line, bb_rect, groundtruthRect);
 170             if (visualize_delay >= 0) cv::rectangle(image, groundtruthRect, CV_RGB(255, 0, 0), 1);
 171             std::cout << ", accuracy: " << accuracy;
 172             sum_accuracy += accuracy;
 173         }
 174
 175         std::cout << std::endl;
 176
 177         if (visualize_delay >= 0) {
 178             cv::Point pt(bb.cx, bb.cy);
 179             cv::Size size(bb.w, bb.h);
 180             cv::RotatedRect rotatedRectangle(pt, size, bb.a);
 181
 182             cv::Point2f vertices[4];
 183             rotatedRectangle.points(vertices);
 184
 185             for (int i = 0; i < 4; i++)
 186                 cv::line(image, vertices[i], vertices[(i + 1) % 4], cv::Scalar(0, 255, 0), 2);
 187             //             cv::rectangle(image, cv::Rect(bb.cx - bb.w/2., bb.cy - bb.h/2., bb.w, bb.h), CV_RGB(0,255,0),
 188             //             2);
 189             std::string angle = std::to_string(bb.a);
 190             angle.erase(angle.find_last_not_of('0') + 1, std::string::npos);
 191             angle.erase(angle.find_last_not_of('.') + 1, std::string::npos);
 192             cv::putText(image, "Frame: " + std::to_string(frames) + " " + angle + " angle",
 193                         cv::Point(0, image.rows - 1), cv::FONT_HERSHEY_SIMPLEX, 0.7, cv::Scalar(0, 255, 0), 2);
 194             cv::imshow("output", image);
 195             int ret = cv::waitKey(visualize_delay);
 196             if (visualize_delay > 0 && ret != -1 && ret != 255) break;
 197         }
 198
 199         //        std::stringstream s;
 200         //        std::string ss;
 201         //        int countTmp = frames;
 202         //        s << "imgs" << "/img" << (countTmp/10000);
 203         //        countTmp = countTmp%10000;
 204         //        s << (countTmp/1000);
 205         //        countTmp = countTmp%1000;
 206         //        s << (countTmp/100);
 207         //        countTmp = countTmp%100;
 208         //        s << (countTmp/10);
 209         //        countTmp = countTmp%10;
 210         //        s << (countTmp);
 211         //        s << ".jpg";
 212         //        s >> ss;
 213         //        //set image output parameters
 214         //        std::vector<int> compression_params;
 215         //        compression_params.push_back(CV_IMWRITE_JPEG_QUALITY);
 216         //        compression_params.push_back(90);
 217         //        cv::imwrite(ss.c_str(), image, compression_params);
 218     }
 219
 220     std::cout << "Average processing speed: " << avg_time / frames << "ms (" << 1. / (avg_time / frames) * 1000 << " fps)";
 221     if (groundtruth_stream.is_open()) {
 222         std::cout << "; Average accuracy: " << sum_accuracy/frames << std::endl;
 223         groundtruth_stream.close();
 224     }
 225     std::cout << std::endl;
 226
 227     return EXIT_SUCCESS;
 228 }