Research On Human Body Gesture Recognition Method Based On Deep Learning

Human gesture recognition,as a research hotspot in the field of computer vision and pattern recognition,has been widely used in many fields such as human-computer interaction,intelligent video surveillance,motion analysis,and medical assistance.In recent years,with the rise of artificial intelligence,the study of human gesture recognition using deep learning methods has attracted more and more scholars' attention,and many human gesture recognition methods and theories based on deep learning have been proposed.However,in the process of human gesture recognition,misjudgments still occur due to intra-class gaps,similar movement confusion,and interference from other external factors.Aiming at the problem of similar motion confusion in the process of human gesture recognition,this paper uses deep learning and data fusion to achieve accurate recognition of similar motions of human body.In the process of solving the problem,the specific work is as follows:(1)Aiming at the problem of similar motion confusion in the human gesture recognition process,this paper uses the methods of data fusion and deep learning to build a DCLSTM(Double CNN LSTM)network model based on CNN and RNN neural networks.(2)The wearable motion capture device is used to collect the kinematic data of the key nodes of the human body,and Openpose is used to extract the coordinate data of the key nodes of the human body in the video.The two types of data are pre-processed to generate pseudo-RGB images and pseudo-grayscale images,and feature extraction,data fusion,and timing analysis are performed on these two kinds of pseudo-pictures.(3)By analyzing the characteristics of two kinds of human gesture data,the feature extraction module,data fusion module and time series module in DCLSTM network model are designed.The loss function is constructed according to the research purpose and network structure,and a suitable classifier and optimizer are selected.(4)The experimental results show that the average accuracy rate of the method in12 poses is more than 96%,and the DCLSTM network model also has a good recognition effect for similar actions of people outside the data set.Validity and generalization.