Abstract: In recent years, certain research progress has been made on suspicious target tracking algorithms for intelligent video surveillance in public places, which is beneficial to public security governance and emergency incident prevention. With the complete attention mechanism in Transformer, we attempt to apply it to the continuous and accurate tracking and positioning of suspicious targets in public security governance. Specifically: Analyze the characteristics of the suspicious target tracking task, explore the novel interaction between the target template framework and the search framework, and propose a deformable transformer tracking algorithm for suspicious targets in public security governance, named DeTrack, which is based on the deformable attention mechanism. The encoder module and the encoder module based on the self-attention mechanism are used, and their combination is used for feature interaction, and then the developed corner prediction head is used to locate the suspicious target. DeTrack can locate targets without paying attention to all pixels, reduces the amount of model parameters, and achieves better tracking performance on LaSOT, TrackingNet, GOT-10K and VOT2020.