Learning Weighted Multi-model Correlation Filters for Visual Tracking

Abstract Correlation Filter (CF) based algorithms play an important role in the field of Visual Object Tracking (VOT) due to its high performance and low computational complexity. While existing CF tracking algorithms suffer performance degradation due to inaccurate object modeling. In this paper, we improved the object modeling accuracy in both CF training stage and target detection procedure, and boosted the overall performance of CF-based trackers. Specifically, we propose a multi-model structure for CF trackers to catch the target appearance change, where different appearance model is trained in a separate frame to catch the saliency character of the target and reduce the computational cost. Furthermore, a space filter for feature maps is designed to suppress the boundary effect under Gaussian motion prior. It contributes to improve the accuracy of position estimation. We deploy our methods to four hand-crafted features based CF trackers to perform real-time visual tracking on popular benchmarks. The experimental results demonstrate the efficacy of our proposed schemes and the efficiency of our trackers. In addition, we have made a comprehensive analysis on the proposed methods for the convenience of reproducing our results and using our schemes.