An RGBD Tracker Based on KCF Adaptively Handling Long-Term Occlusion

摘要

Since occlusion still be a challenge for object tracking in RGB data. In this paper, we propose an RGBD single-object tracker that built upon the well-known base KCF tracker and exploit how the depth information fusing to handle partial and long-term occlusion. To divides tracking model into parts, the proposed tracker could detect and handle occlusion of each part separately. Despite the robustness in tracking with long-term occlusion, our part-based tracker provides an adaptively updating learning matrix. Experimental results are conducted on our dataset, which demonstrate that our tracker contains stability in long-term tracking.