Moving Objects Extraction and Classification Based on Probabilistic Models in Video Surveillance

In order to overcome limitations of traditional clustering algorithms which can't model accurately because of the noise in image data sets, this paper presents an algorithm to segment the current video frame firstly according to spatial properties using the probabilistic model of the Gaussian mixture model of which parameters are estimated by the penalized maximum likelihood estimation, and obtains preliminary outline of moving objects using the block-matching motion estimation algorithm according to temporal properties. Finally, moving objects are extracted if there is a match between the preliminary outline and the segmented image, and classified using the Bayesian probabilistic model.