Supplementary Virtual Keypoints of Weight-Based Correspondences for Occluded Object Tracking

In this paper, we propose a short-term tracking method, which is more robust than conventional methods for single-target tracking under occlusions. On the one hand, our method uses the features-supplemental points and optical flows for static-adaptive targets based on the online tracking algorithm, which clusters the static-adaptive correspondences for deformable object tracking. On the other hand, our method combines the locally matched target bounding box and the one from optical flows tracking using novel fuzzy weights, which are updated based on the changes in keypoints. In the proposed weighted-based keypoints matching tracker, we employ fast keypoints detection and partial descriptors, which makes it run in real time. We demonstrate the effectiveness of our method extensively on benchmark data sets that contain scenes of object deformation and occlusions. The results show that our algorithm performs more favorably against the competing tracking methods.