Saulius' devBlog

I am starting to like the Open Computer Vision Library (OpenCV). I have done some image processing with it for my art class (link) and I really enjoyed it. But when I tried to make some web cam application I realised that OpenCV fails to detect my web camera. After about 2 days spent installing/uninstalling various Microsoft SDK’s I found out that the problem may be somewhere in Visual Studio, because the program crashes only when I start it with debugging, and it works like a charm on ctrl+F5 – no debugging. I couldn’t find solution for this for now.

However, I have still managed to create a sample movement detection application with OpenCV. And this is how it looks:

Ain’t computer science fun?

The algorithm

Well, the algorithm for movement detection here is pretty simple:

1. Get two frames from webcam
2. Blur the images to reduce camera noise
3. Calculate absolute differences from them
4. Make the new image with only those differences
5. Convert step 4 image to grayscale
6. Add binary threshold filter to leave only white and black pixels in grayscale image
7. If there are more than 100 white pixels in area of a circle, then movement is detected!

Implementation

OpenCV provide us very good tools for doing transformations mentioned in algorithm.

Blurring is done with cvSmooth(); calculating absolute difference cvAbsDiff(); converting to  grayscale – cvCvtColor(); threshold – cvThreshold();

A simple snippet to do first 6 steps of algorithm (extracted from the sourcecode at the bottom):

void frameProcessor::compareImages(IplImage * img1, IplImage * img2) //Compare two images (frame)
{
    this->clear(); //Clears image buffers.

	//Check if it's images
	if ((img1 == 0) || (img2 == 0)) { fprintf(stderr, "One/both of images to compare are null\n"); return; }

	//Clone image in order not to modify the original
	this->img1 = cvCloneImage(img1);
	this->img2 = cvCloneImage(img2);

	//Create processed image buffer
	this->processedImg = cvCreateImage(cvGetSize(this->img1), 8, 3); 

    //Blur images to get rid of camera noise
	cvSmooth(this->img1, this->img1, CV_BLUR, 3);
	cvSmooth(this->img2, this->img2, CV_BLUR, 3); 

	//Calc absolute difference
	cvAbsDiff(this->img1, this->img2, this->processedImg);

	//Create gray image buffer
	this->processedImgGray = cvCreateImage(cvGetSize(this->processedImg), 8, 1);

    //Convert colored image to grayscale
	cvCvtColor(this->processedImg, this->processedImgGray, CV_RGB2GRAY);

	//Perform binary treshold filter on image to leave only white and black pixels
	cvThreshold(this->processedImgGray, this->processedImgGray, 30, 255, CV_THRESH_BINARY);
}

And the white color count in area is calculated by following function (also extracted from the source):

int frameProcessor::getWhiteInArea(int startX, int startY, int w, int h)
{
   if (this->processedImgGray == 0) return -1; //If there's no grayscale processed image, return error.

   //Go back to bounds
   if (startX < 0) startX = 0;
   if (startX >= this->processedImgGray->width) return -1;

   if (startY < 0) startY = 0;
   if (startY >= this->processedImgGray->height) return -1;

   //Initialise counter
   int whiteCount = 0;

   //Loop through area
   for (int x = startX; ((x <= startX+w) && (x < this->processedImgGray->width)); x++)
   {
      for (int y = startY; ((y <= startY+h) && (y < this->processedImgGray->height)); y++)
	  {
		  int tmp = ((uchar*)(this->processedImgGray->imageData + this->processedImgGray->widthStep*y))[x];
		  if (tmp == 255) whiteCount++; //If it's white -> add to whitecount.

	  }
   }

   return whiteCount;
}

And that is everything you need for motion detecting with OpenCV. Now you only check if there are more than, let’s say 100 pixels in area and say that the movement was detected here.

The Sourcecode

The application source is available to download here. The additional dependencies for the project are cv.lib highgui.lib cxcore.lib.

Feel free to comment, edit it, etc.

Additional Links

• OpenCV on sourceforge.net
• Andybest.net covering the same topic