Alexey
2017-11-02 49d39922e6c2f991bcc7f446bcde535b35d24a87
src/yolo_v2_class.hpp
@@ -11,9 +11,17 @@
#endif   // OPENCV
#ifdef YOLODLL_EXPORTS
#if defined(_MSC_VER)
#define YOLODLL_API __declspec(dllexport) 
#else
#define YOLODLL_API __attribute__((visibility("default")))
#endif
#else
#if defined(_MSC_VER)
#define YOLODLL_API __declspec(dllimport) 
#else
#define YOLODLL_API
#endif
#endif
struct bbox_t {
@@ -39,30 +47,52 @@
   YOLODLL_API Detector(std::string cfg_filename, std::string weight_filename, int gpu_id = 0);
   YOLODLL_API ~Detector();
   YOLODLL_API std::vector<bbox_t> detect(std::string image_filename, float thresh = 0.2);
   YOLODLL_API std::vector<bbox_t> detect(image_t img, float thresh = 0.2);
   YOLODLL_API std::vector<bbox_t> detect(std::string image_filename, float thresh = 0.2, bool use_mean = false);
   YOLODLL_API std::vector<bbox_t> detect(image_t img, float thresh = 0.2, bool use_mean = false);
   static YOLODLL_API image_t load_image(std::string image_filename);
   static YOLODLL_API void free_image(image_t m);
   YOLODLL_API int get_net_width() const;
   YOLODLL_API int get_net_height() const;
   YOLODLL_API std::vector<bbox_t> tracking(std::vector<bbox_t> cur_bbox_vec, int const frames_story = 6);
#ifdef OPENCV
   std::vector<bbox_t> detect(cv::Mat mat, float thresh = 0.2)
   std::vector<bbox_t> detect(cv::Mat mat, float thresh = 0.2, bool use_mean = false)
   {
      if(mat.data == NULL)
         throw std::runtime_error("file not found");
         throw std::runtime_error("Image is empty");
      auto image_ptr = mat_to_image_resize(mat);
      return detect_resized(*image_ptr, mat.size(), thresh, use_mean);
   }
   std::vector<bbox_t> detect_resized(image_t img, cv::Size init_size, float thresh = 0.2, bool use_mean = false)
   {
      if (img.data == NULL)
         throw std::runtime_error("Image is empty");
      auto detection_boxes = detect(img, thresh, use_mean);
      float wk = (float)init_size.width / img.w, hk = (float)init_size.height / img.h;
      for (auto &i : detection_boxes) i.x *= wk, i.w *= wk, i.y *= hk, i.h *= hk;
      return detection_boxes;
   }
   std::shared_ptr<image_t> mat_to_image_resize(cv::Mat mat) const
   {
      if (mat.data == NULL) return std::shared_ptr<image_t>(NULL);
      cv::Mat det_mat;
      cv::resize(mat, det_mat, cv::Size(get_net_width(), get_net_height()));
      return mat_to_image(det_mat);
   }
   static std::shared_ptr<image_t> mat_to_image(cv::Mat img)
   {
      std::shared_ptr<image_t> image_ptr(new image_t, [](image_t *img) { free_image(*img); delete img; });
      *image_ptr = mat_to_image(mat);
      return detect(*image_ptr, thresh);
      std::shared_ptr<IplImage> ipl_small = std::make_shared<IplImage>(img);
      *image_ptr = ipl_to_image(ipl_small.get());
      rgbgr_image(*image_ptr);
      return image_ptr;
   }
private:
   static image_t mat_to_image(cv::Mat img)
   {
      std::shared_ptr<IplImage> ipl_small = std::make_shared<IplImage>(img);
      image_t im_small = ipl_to_image(ipl_small.get());
      rgbgr_image(im_small);
      return im_small;
   }
   static image_t ipl_to_image(IplImage* src)
   {
      unsigned char *data = (unsigned char *)src->imageData;
@@ -113,57 +143,6 @@
#endif   // OPENCV
   std::deque<std::vector<bbox_t>> prev_bbox_vec_deque;
public:
   std::vector<bbox_t> tracking(std::vector<bbox_t> cur_bbox_vec, int const frames_story = 4)
   {
      bool prev_track_id_present = false;
      for (auto &i : prev_bbox_vec_deque)
         if (i.size() > 0) prev_track_id_present = true;
      static unsigned int track_id = 1;
      if(!prev_track_id_present) {
         //track_id = 1;
         for (size_t i = 0; i < cur_bbox_vec.size(); ++i)
            cur_bbox_vec[i].track_id = track_id++;
         prev_bbox_vec_deque.push_front(cur_bbox_vec);
         if (prev_bbox_vec_deque.size() > frames_story) prev_bbox_vec_deque.pop_back();
         return cur_bbox_vec;
      }
      std::vector<unsigned int> dist_vec(cur_bbox_vec.size(), std::numeric_limits<unsigned int>::max());
      for (auto &prev_bbox_vec : prev_bbox_vec_deque) {
         for (auto &i : prev_bbox_vec) {
            int cur_index = -1;
            for (size_t m = 0; m < cur_bbox_vec.size(); ++m) {
               bbox_t const& k = cur_bbox_vec[m];
               if (i.obj_id == k.obj_id) {
                  unsigned int cur_dist = sqrt(((float)i.x - k.x)*((float)i.x - k.x) + ((float)i.y - k.y)*((float)i.y - k.y));
                  if (cur_dist < 100 && (k.track_id == 0 || dist_vec[m] > cur_dist)) {
                     dist_vec[m] = cur_dist;
                     cur_index = m;
                  }
               }
            }
            bool track_id_absent = !std::any_of(cur_bbox_vec.begin(), cur_bbox_vec.end(), [&](bbox_t const& b) { return b.track_id == i.track_id; });
            if (cur_index >= 0 && track_id_absent)
               cur_bbox_vec[cur_index].track_id = i.track_id;
         }
      }
      for (size_t i = 0; i < cur_bbox_vec.size(); ++i)
         if (cur_bbox_vec[i].track_id == 0)
            cur_bbox_vec[i].track_id = track_id++;
      prev_bbox_vec_deque.push_front(cur_bbox_vec);
      if (prev_bbox_vec_deque.size() > frames_story) prev_bbox_vec_deque.pop_back();
      return cur_bbox_vec;
   }
};