Fast tracking algorithm based on improved particle filter

In recent years, occluded target tracking has become a hot topic in the research field of machine vision. Mean shift(MS) and particle filter(PF) are two successful methods for target tracking. Both have respective advantages and weaknesses. MS algorithm is good at prompting tracking but sensitive to occlusion while PF algorithm is robust to occlusion but need extensive computation cost. In order to achieve the efficient tracking algorithm, a novel improved particle filter was proposed for target tracking in this paper. First, The proposed mean shift forecasted particle filter(MSFPF) improves the sampling efficiency considerably by incorporating MS into PF as a forecasting progress. Meanwhile, the MS vector makes a difference in increasing accuracy of target tracking and reducing computation complexity. The experimental results demonstrated that the proposed algorithm is robust and achieved real-time tracking, the algorithm presented satisfactory performance in full stationary occlusion, the moving completely occlusion with similar color interference as well as the target deformation in an omni-directional vision system.