Research On Human Behavior Monitoring System Of Car Elevator Based On OpenPose

With the development of society,car elevator has become the standard equipment of mid-and high-rise buildings,which is accompanied by the frequent elevator accidents.It is studied that accidents are often caused by uncivilized behaviors during the elevator ride.The traditional elevator monitoring system can only capture and save the video,but cannot give the reminder and warning when the passengers make dangerous actions so as to interrupt them in time.This thesis aims to reduce the accident rate caused by uncivilized behaviors in the car elevator ride through a set of human behavior detection technology that captures and identifies the behaviors with the intelligent monitoring system,giving warning to and interrupting dangerous actions in time.Aiming at the complexity of the traditional image recognition feature extraction,vulnerability of RGB image features to external environment and low recognition rate of complex motions,this thesis shows that the human behavior recognition in the car elevator is achieved via OpenPose algorithm.It studies from four erspectives—single target with various behaviors,double targets with various behaviors,various targets and empty car elevator,within the scope where the abnormal behaviors refer to running,jumping and tumbling and normal behaviors are relatively quiet,such as playing the phone and standing still.The study starts from attaining the original data of human behaviors in three aspects,including extracting the key frames of targeted behaviors in the video stream where the temporal information is obtained to get the sequential information of actions.From the space information of actions is gained from extracting the key nodes of key frame sequences through OpenPose algorithm.The coordinate sequences of key nodes obtained are used to set up a model as data collection which includes the sequences of key nodes describing target actions with temporal and space information.Then the BP neural network in the MATLAB,a neural network toolkit is used to establish a human behavior recognition model whose feasibility is tested through comparing the recognition rates from three perspectives while the positions of the real-time action-capturing cameras of the monitoring system are determined.Finally,the system of real-time monitoring human behavior in the car elevator is designed on the basis of Pyqt5,whose front end and background system and other modules are connected by Python language,so as to realize the capture of human behavior video,the extraction of key frames and the acquisition of key nodes of actions.The information will be imported into the model for recognition and outputting the results.The main function of the system is to capture human behavior in the elevator in real time so as to warn and stop abnormal dangerous actions.The test results of the human behavior monitoring system in the car elevator based on OpenPose proposed in this thesis,show that its recognition rate reaches98.88% and its error rate keeps within 2%.Therefore,it signifies that the model design is of accurate and stability,and the system design is feasible to a certain degree.