Video segmentation based on the Gaussian mixture updating model

Video segmentation is a significant pre-process step in many video analysis systems. In consideration of many current video segmentation methods are time and memory consuming, we present an efficient method in this paper based on the Gaussian mixture model (GMM) with a backward updating model. The Gaussian mixture components produced by the current frame will be used to segment the next frame, and the segmentation result will update the position of each mixture component for the next frame. In this model, color, texture and position features are combined to describe each pixel. Experimental results show this method is fast and robust to the region occlusions.