Abstract: In the field of campus security, the identification of violent behaviors currently mainly relies on manual labor, which is prone to omissions. Skeleton-based spatio-temporal graph convolutional network (ST-GCN) has high behavior recognition accuracy, but it is mainly used for single person recognition. This paper proposes a method of identifying violence against campus surveillance video, which adds a multi-target tracking module on the basis of ST-GCN. The OpenPose algorithm was used to obtain the human skeleton set in the video frame, and the single-person skeleton sequence was separated by the Markov chain Monte Carlo data association method and input into ST-GCN for violent behavior recognition. The experimental results on the data set RWF-2000 show that the recognition rate of this method reaches 87.75%, which is higher than other existing models.