Dangerous Action Recognition for Spatial-Temporal Graph Convolutional Networks

Human behavior recognition is important to research in intelligent surveillance systems. For the criminal act of dangerous accident prevention and identification, dangerous human behavior of recognition is more than important. Aiming at the identification of dangerous human behavior, a human risk behavior detection algorithm based on a time-space graph convolutional network is proposed. The algorithm first extracts the skeleton sequence of human behavior by monitoring the video in the system. Then based on the obtained skeleton features, combined with the time vector in the skeleton sequence, a spatial time map model is established to identify dangerous behaviors. Finally, the hazard level of the behavior is determined according to the setting of the corresponding threshold. Experimental results on the Kinetics and RGB+D datasets show that the algorithm can identify a variety of dangerous behaviors and can effectively identify dangerous behaviors in different scenarios.