Research on behavior recognition method of live working personnel based on human-object interaction detection
In order to solve the problems that the existing video behavior recognition methods cannot distinguish some similar behavior in the process of live working,the number of identifiable types is small,and the interaction between personnel and objects is not utilized,a behavior recognition method of live working personnel in distribution network based on human-object interaction detection was proposed.Firstly,the lightweight pose estimation algorithm was used to identify the human skeleton sequence.Secondly,the spatio-temporal features of human motion were extracted and initially classified by the spatial tempo-ral graph convolutional networks(ST-GCN).Finally,for the similar behavior that cannot be effectively distinguished by skeletal posture,the target detection algorithm was used to identify the tools used by the personnel and the state of use,and the accurate recognition of video behavior was realized by fusing the behavior information contained in human action and oper-ation tools.The results show that the proposed method can effectively identify the live working behavior,and the recognition accuracy of similar behavior is about 88.9％,which is about 53％higher than that of the existing method based on skeleton sequence.The research results can provide reference ideas for improving the level of on-site safety management and control.
live workinghuman-object interactionbehavior recognitionST-GCNskeleton sequence
万方数据
维普
