Siamese network object tracking algorithm based on dynamic template updating

Siamese network tracking methods have demonstrated competitive performance in the domain of single object tracking, with the template updating strategy constituting a pivotal approach to enhancing the siamese network to address issues such as occlusion, deformation, and background clutter. In this paper, an adaptive template updating method based on the dual metrics of maximum response value and APCE value is proposed. The proposed method involves the selection of long and short-term templates through the implementation of a hierarchical template strategy, the establishment of distinct thresholds utilizing the dual metrics, and the dynamic integration of the high-confidence long-term template and the time-sensitive short-term template to obtain template feature information that is more proximate to the tracking object. This approach is intended to ensure the accuracy of the tracking templates and to enhance the performance of the algorithms in dynamic scenarios. Experiments are conducted on the OTB100 dataset, and the experimental results show that the success rate and accuracy of the algorithm in this paper reach 90% and 69.1%, respectively, and the object tracking accuracy is significantly improved in complex situations such as occlusion and deformation.