Learning Temporary Block-Based Bidirectional Incongruity-Aware Correlation Filters for Efficient UAV Object Tracking

In the field of UAV object tracking, correlation filter based approaches have received lots attention due to their computational efficiency. The methods learn filters by ridge regression and generate response maps distinguish specified target from background. An ideal can predict object's position in a new frame, turn, backtrack past frames. However, neglect tracking reversibility most limits potential using inter-frame information improve performance. this work, novel bidirectional incongruity-aware is presented on nature reversibility. proposed method incorporates response-based incongruity, which represents gap between filters' discriminative difference forward backward perspective caused appearance changes. It enables not only inherit discriminability previous but also enhance generalization capability unpredictable variations upcoming Moreover, temporary block-based strategy introduced empower accommodate more drastic changes make effective use information. Comprehensive experiments are conducted three challenging benchmarks, including UAV123@10fps, DTB70, UAVDT. Experimental results indicate that has superior performance compared with other 34 state-of-the-art trackers. Our approach permits real-time at ~46.8 FPS single CPU suitable for online applications.