A block-based MGM for object detection in complex scenes

Accurate motion detection is an important precursor to stable object tracking or recognition. In this paper, a new background model is proposed based on Stauffer's Mixture Gaussians Model (MGM) for the object detecting in complex scenes. The sequences we consider exhibit stationary or dynamic backgrounds, illumination changes, and can have been shot by a moving camera. A block-level background model is established rather than pixel-level in MGM; the background model is adopted for three color channels separately and then intersect them to obtain the final frame. Experiments and comparisons to other motion detection methods on different sequences demonstrate the proposed method has better performance for video analysis in complex scenes, and its effectiveness is proved through quantitative analysis. © 2011 ISSN 2185-2766.