<p>Recently, the Mamba model has been applied to visual tracking. Although it exhibits excellent performance and balances computational efficiency, there are still some shortcomings. As a spatiotemporal property of object tracking with certain trend prediction ability, it has not been integrated into Mamba. However, most Mamba scanning sequences flatten the images into 1D sequences, resulting in insufficient ability to extract spatial structural information. As a tracker, it needs to improve computational efficiency, and its composition complexity is increasing. To address these issues, we propose a new visual tracking method (SDM). Firstly, it combines temporal features to enhance the estimation capability of tracking and positioning by utilizing historical information. Then, it includes a multi-scale backbone structure and deformable Mamba blocks that can dynamically adjust the scanning path, enhancing the ability to capture important features and detail changes. Finally, due to Mamba’s linear time complexity, this tracking method reduces the overall model parameters and improves tracking computational efficiency. To verify the tracking prediction ability of the tracker, we conducted relevant experiments on benchmark datasets such as GOT-10K, LaSOT, TrackingNet and VOT2020. The results indicate that the SDM experiment is effective and competitive.</p>