A Real-time RGB-D tracker based on KCF
Han Zhang, Meng Cai, Jianxun Li · 2018
RGB-D tracking has attracted attentions of researchers in computer vision community, there is still a need to boost processing rate. We propose a RGB-D tracker based on color-only KCF which is able to handle the scale changes and occlusions. The proposed tracker fuses the response maps instead of the features to boost both the accuracy and processing rate. Meanwhile, depth data is utilized to form an indicator which is able to detect occlusions and reappearance of target object effectively. The experimental results evaluated on the public Princeton dataset demonstrate that the proposed tracker achieves state-of-the-art performance in accuracy while runs at speed exceeding 47 frames per second.