A Gaussian mixture based optical flow modeling for object detection

This paper treats the problem of reliable foreground object classification from scene background in an image sequence. Efficient solution for this problem is crucial in the development of automatic video surveillance and tracking systems. We present a background/foreground segmentation approach based on a subtraction of background model that combine color and optic flow information of the scene. A new technique to integrate optical flow information with color information in the background model is developed. The optical flow information complements a color background subtraction model based on spatially global Gaussian mixture. Experimental results that test the proposed approach showed better segmentation than color background/foreground segmentation approach.