Long-term visual tracking based on correlation filters

Abstract

In order to accomplish the long term visual tracking task in complex scenes, solve problems of scale variation, appearance variation and tracking failure, a long term tracking algorithm is given based on the framework of collaborative correlation tracking. Firstly, we integrate several powerful features to boost the represent ability based on the kernel correlation filter, and extend the filter by embedding a scale factor into the kernelized matrix to handle the scale variation. Then, we use the Peak-Sidelobe Ratio to decide whether the object is tracked successfully, and a CUR filter for re-detection the object in case of tracking failure is learnt with random sampling. Corresponding experiment is performed on 17 challenging benchmark video sequences. Compared with the 8 existing state-of-the-art algorithms based on discriminative learning method, the results show that the proposed algorithm improves the tracking performance on several indexes, and is robust to complex scenes for long term visual tracking.