Adaptive Hyper-Feature Fusion for Visual Tracking

Zhi Chen, Yongzhao Du, Jianhua Deng, Jiafu Zhuang, Peizhong Liu

Abstract: In this work, we propose a robust tracking algorithm based on context-aware correlation filter framework. In order to improve the richness of the feature representation, the proposed hyper-feature which contains linearly weighted mixture of hand-crafted features (such as HOG, color histogram) and hierarchical deep convolutional features (such as VGGNet). The final output response map is optimized by the Gaussian constrained optimization method to control the response map follow the Gaussian distribution, which gain the robustness to target appearance variations. In addition, in terms of model update, an efficient adaptive model updating method is proposed to suppress the model noises significantly. Extensive experimental results on well-known tracking benchmark datasets to evaluate the proposed algorithm. Experimental results demonstrate that the proposed algorithm performs favorably against many state-of-the-art methods in terms of success rate, accuracy, and robustness.