A Novel Object Tracking Algorithm Based On Enhanced Perception Hash And Online Template Matching

Jin Yuan,Dong Xu,Heng-Chang Xiong,Zhi-Yong Li

DOI: https://doi.org/10.1109/FSKD.2016.7603223

2016-01-01

Abstract:Object tracking task faces serious challenges when a desired target is in complex circumstances such as obstacle occlusion, pose variation, illumination change and motion blur. Despite lots of excellent tracking algorithms have been proposed, many issues remain to be addressed. In this paper, we propose a novel object tracking algorithm to combine both Enhanced Perception Hash and Coarse-to-fine Sliding Window search strategy. First, we calculate the feature template of target by using the perception hash approach integrating Fast Fourier Transform (FFT). We make a FFT on the target area in a frame and only retain low-frequency part to save storage. Moreover, the new template is generated by fusing the templates from the current and previous frames, thus it is robust to the severe deformation and drastic variation problem. Second, we propose a coarse-to-fine sliding window search strategy to provide potential target candidates efficiently. Based on this strategy, our tracking algorithm can implement an effective online detection for target. Extensive tests on thirteen videos are conducted and demonstrate that our approach achieves promising performance as compared to the state-of-the-art methods.