Adaptive foreground segmentation using fuzzy approach

Abstract

In this paper, we propose a simple and novel method for background modeling and foreground segmentation for visual surveillance applications. This method employs histogram based median method using HSV color space and a fuzzy k-means clustering. A histogram for each pixel among the training frames is constructed first, then the highest bin of the histogram is chosen and the median value among this bin is selected as the estimated value of background model for this pixel. A background model is established after the above procedure is applied to all the pixels. Fuzzy k-means clustering is used to classify each pixel in current frame either as the background pixel or the foreground pixel. Experimental results on a set of indoor videos show the effectiveness of the proposed method. Compared with other two contemporary methods — k-means clustering and Mixture of Gaussians (MoG) — the proposed method is not only time efficient but also provides better segmentation results.