Visual object tracking using Kalman filter, mean shift algorithm and spatiotemporal oriented energy features

Many multimedia applications need to track moving objects. Consequently, designing a robust tracking system is a vital requirement for them. This paper proposes a new method for visual object tracking, which uses the mean shift tracking algorithm to derive the most similar target candidate to the target model. Bhattacharyya coefficient is employed to determine the similarities. Target's structure is represented by multiscale oriented energy feature set, which presents extra robustness by including dynamic information of the pixels. Likewise, the Kalman filtering framework is employed to predict the location of the moving objects. Experimental results demonstrate the proposed algorithm's superior performance, chiefly when encountering with the full occlusion situation.