Multi-Experts with Context Awareness Correlation Filters for Visual Tracking
Qixiang Zou, Da Li, Xuejun Chen · 2020
Correlation filters based trackers have achieved competitive performance in tracking community due to their high computational efficiency in Fourier domain. However, a large number of recent research focuses on improving the representative power of the target, which is still difficult to distinguish between target and background. At the same time through fusion and the use of deep features, the computational burden will be increased. In this paper, a multi experts with context aware correlation filters framework is proposed. Under this framework, target and contextual information are extracted to construct experts, and a most reliable one is selected as the result of current frame. Through this refining strategy, tracking drift can be alleviated. Extensive experiments are conducted on three benchmarks demonstrate that the proposed trackers performs favorably against state-of-the-art approaches.