The Design And Implementation Of A Dangerous Behavior Recognition System For Substation Monitoring

In the substation monitoring scene,some staff will view and repair the equipment,including aerial work,and these jobs are very dangerous.Therefore,the construction personnel are required to obey the maintenance rules.Once a dangerous behavior occurs,the construction personnel’s life may be threatened.However,checking the occurrence of an event by manually scheduling a video is labor-intensive and inefficient.Therefore,video-based behavior recognition has important research significance.Based on the national grid project,this paper mainly studies the problem of human dangerous behavior recognition in substation monitoring scenarios,and designs the recognition system.The scenes include high-altitude scenes and fall behavior scenes.Among them,the dangerous behavior in the high-altitude scene is mainly the irregular posture of the maintenance personnel;and the fall behavior scene is the fall behavior as a dangerous behavior,which mainly includes the action of the fall and the state after the fall.The processes include: First,human detection.We adjusted and refined candidate box continuously through Fast-RCNN,and finally obtained the position of the human body.Second,limb key point mapping.We established a skeleton model,and located the key points of limbs through the Alpha Pose model,and got coordinate information to achieve mapping.Thirdly,the dangerous behavior recognition.We have proposed the dangerous behavior recognition method based on the temporal and spatial feature of key point.This method transforms the dangerous behavior semantics of different scenes into corresponding key point feature descriptions.Once dangerous behaviors are detected,the system will give corresponding alarm information prompts.The research of this thesis has strong application ability for behavior analysis in substation scenarios,which can smoothly promote the implementation of national grid projects,and also has a good inspiration for behavior analysis in other security fields.