Background Modeling Using Time Dependent Markov Random Field With Image Pyramid

Background modeling is important for video surveillance. In the paper, we present a novel background modeling algorithm based on probabilistic graphic model and Gibbs Sampling. The background is modeled at different resolution level by image pyramids. We develop a time dependent pyramidal Markov Random Field (MRF) to represent the state of foreground/background at each pixel in the pyramid. Both spatial and temporal constraints in the foreground labeling process are considered using three kinds of links in the MRF. The probability of adding/deleting a foreground object is calculated by online learning algorithm and is used as prior information in computing foreground label. We use Gibbs Sampling to solve the MRF under the framework of Maximum A Posterior (MAP). Experimental result shows that this real-time algorithm is able to segment the foreground object accurately from videos with clutter in the scene.