Target re-location kernel correlation filtered visual tracking with fused deep feature

To enhance the tracking robustness of the kernel correlation filtering algorithm in complex scenarios, the present study combines the traditional kernel correlation filtering algorithm with depth features and proposes a correlation filtering algorithm incorporating depth features. At first, HOG features, CN features and depth features are extracted and fused, while the dimensionality of the features is reduced separately with the aim to reduce the computational effort. Secondly, a scale pool is established, and the scale search method is employed to improve the computational efficiency of scale estimation. Thirdly, an active detection mechanism for occlusion is introduced, which uses different methods for anti-occlusion processing by classifying the degree of occlusion. Finally, the tracking confidence of current and historical frames is calculated and a threshold is set for adaptive model updating. We have conducted comparison experiments on the OTB-100, VOT2018, LaSOT, and UAV123 datasets, respectively. The experimental results demonstrate that the algorithm in this paper possesses good robustness in complex scenes.