Exploiting depth information to increase object tracking robustness

Object tracking is a critical component of many computer vision applications. However, despite years of research, it is still considered a difficult problem as most tracking algorithms fail when the objects of interest are similar in appearance to their surroundings or they are occluded. In this paper the widely used 2D mean-shift tracking algorithm is extended to 3D space, in order to exploit the extra depth information provided by 3D video content to increase tracking accuracy and robustness. The performance of the proposed 3D tracking algorithm is compared to the traditional 2D mean-shift algorithm in terms of tracking accuracy and speed. Results show that in general the tracking accuracy is improved while requiring more processing time. Nonetheless, the algorithm executes at a rate which is faster than typical video capturing rates and thus has no impact on the performance of the tracking system.