An improved background subtraction method based on region growing

In this paper, we propose a simple and effective method for human detection based on region growing to improve background subtraction. In some complex scenes, a person will be detected more than one contour region in such cases as that the colors of person clothes are close to the background color so that the original background subtraction cannot obtain good performance. We use region growing method to cluster the foreground regions which obtained in the scene. After clustering, for each class, it is a person in the image if the sum area of contours exceeds a threshold. Experimental results show that the performance of this method is much better than the original background subtraction and it can detect people very fast.