Bidirectional Joint Attention Mechanism for Target Tracking Algorithm

Abstract

In this paper, we propose a target tracking algorithm SiamMT based on Siamese network architecture. The attention mechanism is introduced to solve the problem of semantic information loss caused by the classical traditional Siamese network architecture. Aiming at the problem that the target feature is not prominent and comprehensive, a bidirectional feature enhancement network model is proposed. The local and global attention mechanisms of MobileNet and Transformer are used to enhance and update bi-directional features, and then the attention mechanism and classification regression network are combined to form a fusion prediction network. This fusion mechanism effectively integrates local and global features, and breaks the limitations of the search domain method by making full use of spatial information and motion information. In order to solve the problem of complexity and inefficiency of tracking network, lightweight network MobileNet is introduced as an enhanced network with local features. Finally, the experiments on OTBIOO and LaSOT long-term benchmark show that the tracker in this paper has higher accuracy and success than other advanced trackers.