Moving objects detection by Gaussian Mixture Model: A comparative analysis

Robust and real-time moving objects detection is a critical issue in computer vision application. The Gaussian Mixture Model (GMM) is the most common method to build a background. In this paper, three GMM algorithms, the classical one, L-window and the algorithm proposed by P. Wayne and Johan, are discussed, and videos of variety scenes are used in the comparison between their performances. Our study results show that 1) The classical algorithm shows satisfactory performance for each scene except fewer holes in detected foreground and does not detect nearly unmoving foreground as the part of background, 2) L-window algorithm is suitable for outdoor high-speed moving object and slowly moving object while the detected foreground is fuzzy with many small blobs, it is unsuitable for illumination changing scene, and 3) The algorithm proposed by P. Wayne and Johan is the best one for indoor sequence within the three algorithms while exhibiting unsatisfactory in case of outdoor scene and slowly moving objects.