GMM-based efficient foreground detection with adaptive region update
Hao Li, Alin Achim, David Bull · 2009
The accurate detection of moving objects is an important step in the process of tracking and recognition in many real-time video surveillance applications. In this paper, we propose a combination of block-based detection and a pixel-based Gaussian Mixture Model (GMM) for moving object detection. Compared with traditional pixel-based algorithms which update all pixels for every frame, our algorithm has the ability to selectively update region information within each frame, while offering the capability to refine the silhouette of a foreground object. The algorithm offers an efficient trade-off between complexity and detection performance. The results show improved detection in the presence of high camera noise, high level compression artefacts, camera movements and dynamic background conditions.