Multiple targets tracking with robust PCA-based background subtraction and Mean-Shift driven Particle filter

Background subtraction, and tracking are the first processing steps in video surveillance applications. Prior work on background modeling has mainly focused on model representation for each pixel, further more, it also need longer training sequences in which no much more moving objects. Here, we introduce a new background modeling approach based on R-PCA for the background subtraction, which can obtain better results from short video contains lots of motions. After that, we use morphological operations to filter out the motion noise, and then combine appearance-based Mean Shift and particle filter tracker for accurately moving target tracking. The proposed algorithm is not only avoiding the degeneracy problem in the particle filter, and also overcoming lost track problem when occlusion occurs. Experiments on real videos captured at different situation shows the effectiveness of the proposed system.