Enhanced real-time mean shift tracking by addressing object occlusion in IR imaging

Although combining Kalman filter (KF) and mean-shift algorithm improves the performance of tracking process, it still malfunctions in cases like large movement or object partial occlusion. Having a little similarity measure between a target model and a target candidate forces the algorithm to stay in initial location and it gets stuck in local maximums and eventually looses desired objects. This paper escapes from local maximum by invoking object detection which locate closest object and choose efficient window size in compare to fixed window size in mean shift. It separates object from IR image background using morphological operators on output of KF. Moreover, this paper applies new nonlinear weighting strategy to make histogram of frames weight-based. Simulation results show that this approach outperforms others in terms of number of iterations and robustness.