Research On Application Of Human Body Recognition In Railway Safety Protection

With the rapid development of artificial intelligence in recent years and the conti nuous updating of hardware technology,especially the computing power of parallel co mputing hardware GPUs has increased rapidly.Image recognition technology has been deeply applied to all aspects of our production and life.Although there are many inn ovations in the human pose image recognition algorithm and good performance,there are still many needs to apply it to the safety protection of railway shunting due to th e complex and changeable scene environment and numerous uncertain factors.Improve ment and innovation aspects.In this paper,based on the previous application of human body pose recognition algorithm and safety protection,this paper uses human bone joint points for feature de scription,and proposes a human joint point pose recognition model for static and dyn amic poses.The distance method and coordinate method are proposed respectively.Sta tic pose recognition is used for the skeleton of a human body in a single frame.The target joint point is calculated as a reference node,and the distance between it and ot her target nodes is converted to construct a feature matrix.Dynamic pose recognition is used for the human skeleton in continuous pictures,also called optical flow informa tion.The displacement of each node relative to the last moment is calculated and use d as the feature vector of the dynamic pose.Experiments show that the accuracy rate of human pose image recognition reaches 96.2%,and the recognition speed can reach 26 frames per second.Applying the human body posture recognition algorithm designed in this thesis,th e railway shunting safety protection monitoring and early warning system was develop ed to address the pain points of railway shunting safety protection accidents.Based on the behavior monitoring and early warning mechanism of railway operators and the v isualization requirements of front-end results,the system proposes a translation selectio n box mechanism for the recognition of human poses and images for the purpose of speed,data cache window and result cache window.Aiming at the accuracy of early warning of abnormal behaviors,an abnormal translation selection box was proposed.B y accumulating the abnormal values of illegal operation behaviors and setting the thres hold of the abnormal values,it was judged whether the alarm and the level of the al arm should be triggered.Finally,the client software is designed and deployed and test ed in the field environment.The human-computer interaction interface is friendly,whic h can realize the real-time safety protection of the shunting operation on site.The ala rm mechanism and snap storage function are designed for abnormal situations.