| The training of command signals for hoisting operation is a requisite part of pre-job training for hoisting operations,and it is also a requirement of safety production.Traditional training methods have high requirements for time,site and personnel,so it is urgent to develop a hoisting operation command signal training system that has less requirement on the above and can meet the training requirements.With the development of computer technology,it has become a new training method to identify and train the act of command signal for hoisting operations using human body posture recognition technology based on deep learning.Based on the training of the command signals for hoisting operation,this paper studies the human posture recognition technology,and applies this technology to design and develop a hoisting operation command signal training system.The specific research content are as follows:(1)Aiming at the problem that the current human pose estimation algorithm performs poorly in estimation accuracy and speed,and the model parameters are too large,the human pose estimation technology is studied.This paper optimized the bottom-up human pose estimation algorithm OpenPose.The feature extraction network is replaced to reduce the amount of model parameters and improve the estimation speed.On this basis,the attention mechanism was introduced to improve the estimation accuracy.The convolution mode of the branch network was optimized to further reduce the number of parameters.Experiments on the COCO dataset show that the estimation accuracy and speed of the optimized algorithm are greatly improved compared with OpenPose,and it also has different degrees of optimization compared with other classic human pose estimation algorithms.(2)Aiming at the problem that the information extraction of temporal features and spatial features cannot achieve favorable results at the same time in the existing human action recognition techniques,a human action recognition network based on human pose estimation(LSTM-GCN)is proposed.By fusing long short-term memory network(LSTM)and graph convolutional network(GCN),their information extraction capabilities in time and space dimensions were respectively played.Based on the JHMDB data set and the self-made lifting operation command signal data set,a time series model of human skeleton key points are constructed as the input data of LSTM-GCN for experimental analysis.The results show that the recognition effect of LSTM-GCN network with human action is in line with expectations.(3)Aiming at the issue of insufficient student immersion and delayed feedback of training effect in traditional training methods of hoisting operation command signal,a hoisting operation command signal training system based on human posture recognition technology is constructed.The system includes training and assessment functions,and designs an action scoring method that takes the Angle deviation value of joint parts between the standard signal posture and the signal posture performed by the student as the evaluation index,which can score the training effect in real time.The system meets the training requirements of hoisting operation command signal,and can do trainers and trainees a favor to achieve the desired training effect.(4)Aiming at the application problem of hoisting operation command signal training system in practical training,the technical background of the training is analyzed,and the training process is described in detail.At the same time,the training effect is analyzed combined with the application scene.The results show that the effect of the system in the hoisting operation command signal training is better than the traditional training method,and it is suitable for promotion in the training of hoisting operation command signal. |
