cvRunningAvgを使って背景統計を取り背景差分する

背景差分法で背景と物体を分離するコードで、opencv.jp - OpenCV: 背景統計量の累積（Accumulation of Background Statistics）サンプルコード -が大変すばらしかったので自分用に使いやすいようにいじくりまわした。

#include <cv.h>
#include <highgui.h>
#include <ctype.h>
#include <stdio.h>

int main(int argc, char **argv)
{
  bool isStop = false;
  const int INIT_TIME = 50;
  const double BG_RATIO = 0.02; // 背景領域更新レート
  const double OBJ_RATIO = 0.005; // 物体領域更新レート
  const double Zeta = 10.0;
  IplImage *img = NULL;

  CvCapture *capture = NULL;
  capture = cvCreateCameraCapture(0);
  //capture = cvCaptureFromAVI("test.avi");
  if(capture == NULL){
    printf("capture device not found!!");
    return -1;
  }

  img = cvQueryFrame(capture);
  int w = img->width;
  int h = img->height;

  IplImage *imgAverage = cvCreateImage(cvSize(w, h), IPL_DEPTH_32F, 3);
  IplImage *imgSgm = cvCreateImage(cvSize(w, h), IPL_DEPTH_32F, 3);
  IplImage *imgTmp = cvCreateImage(cvSize(w, h), IPL_DEPTH_32F, 3);
  IplImage *img_lower = cvCreateImage(cvSize(w, h), IPL_DEPTH_32F, 3);
  IplImage *img_upper = cvCreateImage(cvSize(w, h), IPL_DEPTH_32F, 3);
  IplImage *imgSilhouette = cvCreateImage(cvSize(w, h), IPL_DEPTH_8U, 1);
  IplImage *imgResult = cvCreateImage(cvSize(w, h), IPL_DEPTH_8U, 3);

  printf("背景初期化中...\n");
  cvSetZero(imgAverage);
  for(int i = 0; i < INIT_TIME; i++){
    img = cvQueryFrame(capture);
    cvAcc(img, imgAverage);
    printf("輝度平均 %d/%d\n", i, INIT_TIME);
  }
  cvConvertScale(imgAverage, imgAverage, 1.0 / INIT_TIME);
  cvSetZero(imgSgm);
  for(int i = 0; i < INIT_TIME; i++){
    img = cvQueryFrame(capture);
    cvConvert(img, imgTmp);
    cvSub(imgTmp, imgAverage, imgTmp);
    cvPow(imgTmp, imgTmp, 2.0);
    cvConvertScale(imgTmp, imgTmp, 2.0);
    cvPow(imgTmp, imgTmp, 0.5);
    cvAcc(imgTmp, imgSgm);
    printf("輝度振幅 %d/%d\n", i, INIT_TIME);
  }
  cvConvertScale(imgSgm, imgSgm, 1.0 / INIT_TIME);
  printf("背景初期化完了\n");

  char winNameCapture[] = "Capture";
  char winNameSilhouette[] = "Silhouette";
  cvNamedWindow(winNameCapture, CV_WINDOW_AUTOSIZE);
  cvNamedWindow(winNameSilhouette, CV_WINDOW_AUTOSIZE);

  while(1){
    if(!isStop){
      img = cvQueryFrame(capture);
      if(img == NULL) break;
      cvConvert(img, imgTmp);

      // 輝度範囲
      cvSub(imgAverage, imgSgm, img_lower);
      cvSubS(img_lower, cvScalarAll(Zeta), img_lower);
      cvAdd(imgAverage, imgSgm, img_upper);
      cvAddS(img_upper, cvScalarAll(Zeta), img_upper);
      cvInRange(imgTmp, img_lower, img_upper, imgSilhouette);

      // 輝度振幅
      cvSub(imgTmp, imgAverage, imgTmp);
      cvPow(imgTmp, imgTmp, 2.0);
      cvConvertScale(imgTmp, imgTmp, 2.0);
      cvPow(imgTmp, imgTmp, 0.5);

      // 背景領域を更新
      cvRunningAvg(img, imgAverage, BG_RATIO, imgSilhouette);
      cvRunningAvg(imgTmp, imgSgm, BG_RATIO, imgSilhouette);

      // 物体領域を更新
      cvNot(imgSilhouette, imgSilhouette);
      cvRunningAvg(imgTmp, imgSgm, OBJ_RATIO, imgSilhouette);

      cvErode(imgSilhouette, imgSilhouette, NULL, 1); // 収縮
      cvDilate(imgSilhouette, imgSilhouette, NULL, 2); // 膨張
      cvErode(imgSilhouette, imgSilhouette, NULL, 1); // 収縮

      cvMerge(imgSilhouette, imgSilhouette, imgSilhouette, NULL, imgResult);
      cvShowImage(winNameCapture, img);
      cvShowImage(winNameSilhouette, imgResult);
    }
    int waitKey = cvWaitKey(33);
    if(waitKey == 'q') break;
    if(waitKey == ' '){
      isStop = !isStop;
      if(isStop) printf("stop\n");
      else printf("start\n");
    }
  }

  cvReleaseCapture(&capture);
  cvDestroyWindow(winNameCapture);
  cvDestroyWindow(winNameSilhouette);

  return 0;
}