Advances in object tracking algorithm based on siamese network

Object tracking, as a key topic in the field of computer vision, is widely used in fields such as intelligent video surveillance. With the rapid development of deep learning, the siamese neural network-based tracking algorithm (referred to as siamese tracking algorithm) becomes the mainstream algorithm due to its balanced advantages of speed and accuracy. Despite a large number of studies, there is still a lack of systematic analysis of siamese tracking algorithms from the level of the tracking framework. In order to sort out the current research progress of siamese tracking algorithms, the common challenges, main components, tracking process, common datasets and evaluation indexes of siamese tracking algorithms are firstly introduced. Secondly, the algorithms are divided into algorithms for improving feature extraction, algorithms for optimizing similarity calculation, and algorithms for optimizing tracking results according to the improvement direction of the tracking framework, and they are introduced in detail respectively. Then 20 mainstream tracking algorithms are tested and analyzed. Finally, we summarize the problems of current siamese tracking algorithms and future research directions. © 2022, Editorial Office of Systems Engineering and Electronics. All right reserved.