Hybrid Attention Mechanism combined with Peak Sampling for Object Tracking

In recent years, deep learning has created a research boom in the field of computer vision and continues to drive the development of high-level tasks such as object tracking and object detection. However, designing a robust object tracking method remains a challenging topic due to the tracking challenges such as illumination changes, occlusions, deformations, etc. In this paper, we propose a novel object tracking method that combines attention mechanism and peak sampling to address the problems that arise in most existing object trackers when facing the above challenges. First, an effective hybrid attention mechanism is proposed and then introduced into the model to enhance the tracker's attention to foreground targets. Then, the Maximum peak sampling method is adopted to strengthen the highest peak of the feature response map to highlight the foreground target, which effectively suppresses the interference of target analogues. Finally, a combination of offline training and online learning methods is used to further improve the tracking accuracy and robustness. Experimental results on several standard datasets show that the proposed tracker is able to effectively promote the tracking accuracy and robustness.