Part-Based Co-Difference Object Tracking Algorithm for Infrared Videos

This paper proposes a novel visual object tracking method for IR images extending the previous co-difference feature-based tracking method. The proven success of the co-difference feature based method is further improved by addressing the performance loss in the rotation and appearance change cases. The faster and more significant changes in such cases are handled by dividing the object template into smaller parts which are sampled at salient corner points for robustness. The track patches are updated as in the original method and resampled whenever they are drifted to be outliers. The performance of the proposed method is evaluated on an extensive public thermal image dataset consisting of various surveillance, reconnaissance and targeting scenarios.