Research On Abnormal Sitting Posture Recognition Method Based On Deep Learnin

With the development of society and the improvement of scientific and technological level,people’s demands for health are becoming higher and higher when material conditions are met.Sitting posture,as one of the most common human postures in daily life,has received increasing attention.Maintaining abnormal sitting posture for a long time for various activities often brings many health problems and affects the normal life of people.How to accurately recognize human abnormal sitting posture has become one of the hot research topics.This paper first constructs a dataset of abnormal sitting postures containing multidimensional features from the perspective of the whole human body.Secondly,in order to improve the accuracy of recognizing human abnormal sitting posture,an abnormal sitting posture recognition method based on the multi-scale spatiotemporal features of skeleton graph is proposed.Finally,based on the abnormal sitting posture recognition method proposed in this paper,a human abnormal sitting posture recognition and detection system is designed and developed.The specific research contents are as follows:1)To build a human abnormal sitting posture dataset with complete features and solve the problem of missing features in abnormal sitting posture data in previous work.Firstly,from the perspective of the whole human body,this paper uses a camera to collect video stream data of human sitting posture from normal to abnormal.Secondly,the human posture estimation method is used to obtain spatiotemporal sequence data of skeleton keypoints for abnormal sitting posture.Based on the human skeleton keypoints data,the cosine angle calculation method is used to obtain the human local skeletal angle sequence data.Finally,a human abnormal sitting posture dataset containing multidimensional features is constructed,which includes 6 common types of human abnormal sitting postures and a total of 700 samples.2)In order to improve the accuracy of recognizing human abnormal sitting posture,an abnormal sitting posture recognition method based on the multi-scale spatiotemporal features of skeleton graph(ASPR)is proposed.Firstly,in order to reduce the loss of abnormal sitting posture features during the recognition process,according to the high-tolow resolution subnetworks structure of HRNet,using the spatiotemporal graph convolution module as the feature extraction unit,we construct a multiscale spatiotemporal feature extraction model based on graph convolution neural network(M2SGCN)to extract the spatiotemporal sequence features of skeleton keypoints in human abnormal sitting posture.Then,to extract the features of local skeletal angle sequences in human abnormal sitting posture,a feature extraction model of local skeletal angles based on recurrent neural network is constructed.Finally,in order to further improve the recognition effect of human abnormal sitting posture,a linear weighted feature fusion method is adopted to fuse the features extracted from the above two models to complete the recognition of human abnormal sitting posture.We investigate the optimal parameters of ASPR by checking its performance with different human skeleton combination schemes and different features fusion coefficient.Experiment results show that ASPR achieves excellent performance compared with four state-of-the-art models and their conbined models.ASPR shows an average recognition accuracy,a recall rate,an F1 score,and average time of 95.24%,95.61%,95.14%,and7.094 ms,respectively.3)Based on the human abnormal sitting posture recognition method proposed in this paper,the model is deployed to the PC terminal.After providing the overall system architecture design,functional module construction,and software and hardware integration,a human abnormal sitting posture recognition system is constructed using Python language and web front-end construction tools.