Pedestrian Tracking Based on HSV Color Features and Reconstruction by Contributions

摘要

It is a challenging task to track pedestrian accurately in complicated environment such as illumination, background variation, object occlusion, scale variation, noise, and fast motion. Aiming at these problems, the tracking algorithm based on HSV color features and reconstruction by contributions is proposed. The proposed algorithm extracts the mixed color features of target in HSV space to generate the target template set within the particle filter framework. According to the influence of different regions on the tracking results, the contribution of the region is distributed. And it is introduced into the adaptive regularization model, the region with the minimum reconstruction error is determined as the target to be tracked. In order to be more robust, the templates are updated in real time during the tracking progress. Experiments on OTB100 sequences shows that the proposed algorithm can realize the continuous tracking in complex video scenes and is beneficial to be applied to the practice as a result of its better robustness. The average center error of tracking results and tracking success rate are 0.6624 pixel and 0.4153.