Identification of Child Physical Abuse Based on Openpose and ST-GCN

In order to safeguard the physical and mental health of children, this article proposes a behavior recognition method based on the spatial temporal graph convolutional neural network using skeleton data after customizing mild, moderate, and severe physical abuse behaviors. OpenPose algorithm is introduced to extract human skeleton data for training and recognition tasks, improving the robustness of the model in complex environments. Experimental results show that the recognition accuracy of OpenPose+ST-GCN for mild physical abuse behavior reaches up to 98%, and the average accuracy for recognizing all three levels of physical abuse behavior is 64.13%. The average recognition accuracy for recognizing the three levels of physical abuse behavior using OpenPose+2s-AGCN is 65.52%, which is 1.39% higher than that of OpenPose+ST-GCN.