RGB-T Tracking via Multi-expert Correlation Filters using Spatial-temporal Robustness

RGB-T tracking has made impressive progress due to the complementarity of RGB images and TIR images. However, most CF based trackers use the cross-modal information through simple feature concatenation, which does not take into account the cross-modality discrepancy. In this paper, we propose a multi-expert tracking framework based on discriminative correlation filter. Specifically, a simple robustness evaluation strategy is used to select the most proper tracking result from multiple experts as the tracking result. Besides, we use principal component analysis to reduce redundancy and improve tracking speed, with little sacrifice on the tracking accuracy. Comprehensive experiments on RGBT210 dataset has proved the effectiveness of the proposed tracker.