Block matching-based background generation and non-rigid shape tracking for video surveillance

This paper presents a robust, real-time block matching-based tracking method for deformable objects. Block matching has been used for various image processing applications as well as video coding because it has a regular computing structure. Video tracking based on block matching, however, fails to provide correct motion information of the object when; (i) background changes spatio-temporally, (ii) an object deforms, and (iii) occlusion occurs. To solve these problems, background is first generated to localize a deformable object, and the object is defined by the corresponding boundary. Shape of the object is reconstructed by using a set of shape control points (SCPs) inside the object boundary, and the trajectory of each SCP is traced by block matching. We use an additional SCP Preserving step at every certain interval of frames for more efficient object tracking. For the experiment we used a computer-generated moving fish sequence and an outdoor sequence. Experimental results show that the proposed tracking algorithm can solve the problem of existing block matching-based methods, and at the same time it can track an object with occlusion and/or deformation.