Latency fix for RTSP-camera in console-app

AlexeyAB

2017-08-07 b202eaee259572cf8b627a776a62ce9d59aa50a6

 src/yolo_v2_class.hpp

@@ -1,6 +1,8 @@
#pragma once
#include <memory>
#include <vector>
#include <deque>
#include <algorithm>

#ifdef OPENCV
#include <opencv2/opencv.hpp>       // C++
@@ -9,15 +11,24 @@
#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 {
   unsigned int x, y, w, h;   // (x,y) - top-left corner, (w, h) - width & height of bounded box
   float prob;             // confidence - probability that the object was found correctly
   unsigned int obj_id;    // class of object - from range [0, classes-1]
   unsigned int track_id;     // tracking id for video (0 - untracked, 1 - inf - tracked object)
};

struct image_t {
@@ -31,34 +42,44 @@
class Detector {
   std::shared_ptr<void> detector_gpu_ptr;
public:
   float nms = .4;

   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();
   YOLODLL_API int get_net_height();

   YOLODLL_API std::vector<bbox_t> tracking(std::vector<bbox_t> cur_bbox_vec, int const frames_story = 4);

#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");
      cv::Mat det_mat;
      cv::resize(mat, det_mat, cv::Size(get_net_width(), get_net_height()));
      auto image_ptr = mat_to_image(det_mat);
      auto detection_boxes = detect(*image_ptr, thresh, use_mean);
      float wk = (float)mat.cols / det_mat.cols, hk = (float)mat.rows / det_mat.rows;
      for (auto &i : detection_boxes) i.x*=wk, i.w*= wk, i.y*=hk, i.h*=hk;
      return detection_boxes;
   }

   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;
@@ -107,6 +128,8 @@
   }

#endif   // OPENCV

   std::deque<std::vector<bbox_t>> prev_bbox_vec_deque;
};