Siamese Network Tracking Based on Feature Enhancement
Dandan Huang,Mingting Yang,Jin Duan,Siyu Yu,Zhi Liu
DOI: https://doi.org/10.1109/access.2023.3266264
IF: 3.9
2023-01-01
IEEE Access
Abstract:Most popular Siamese network trackers implement similarity learning by convolutional intercorrelation using the output features of the last layer of the feature extraction network. The extracted image features, although having high semantic information, lack edge as well as some detail information, which is not conducive to meet some challenges in tracking tasks. In this paper, a simple Siamese network tracking based on feature enhancement is proposed to address the above problems. The method applies the output features of the last three layers of the feature extraction network together, which are used to supplement some detailed information and to augment the target information features by a self-attentive mechanism. At the same time, a classification enhancement module is designed in the back-end classification Section to assist in classifying target information more accurately. Finally, this method is experimented on four challenging datasets, namely GOT-10k, UAV123, OTB100 and LaSOT, and the experiments show that the proposed method in this paper outperforms some existing state-of-the-art target trackers. And the method in this paper weighs the tracking accuracy and speed in the model structure design, which both improves the tracker tracking performance and can run at a speed of about 61 fps on such large data as GOT-10k.
computer science, information systems,telecommunications,engineering, electrical & electronic