A robust tracking algorithm with on online detector and high-confidence updating strategy

摘要

The discriminative correlation filter-based tracking algorithms cannot correctly track the target if the target is occluded or out of view and reappears in the field of vision, and they cannot ensure the tracking model is updated correctly if the tracking information is not correct. In this paper, a robust correlation tracking algorithm is proposed. Here, a failure detection strategy, which is based on the maximal confidence score and peak-to-sidelobe ratio to detect or measure the reliability of the tracking result, is integrated into the tracker. Moreover, the redetection module based on the keypoints matching method for consensus voting is introduced into the proposed tracking algorithm to redetect objects in case of tracking failure. In addition, an adaptive high-confidence updating method is proposed to avoid error model information introduced into the tracker caused by occlusions, out-of-view or illumination changes, where the learning rate is determined by the change rate of the confidence map. The OTB-2015 dataset and VOT-2016 dataset are used to evaluate the performance of the proposed tracking algorithm. The experimental results show that the proposed tracking algorithm performs better than most of the state-of-the-art trackers, and it has higher accuracy and robustness than the DSST tracker.