Research Of Motion State Recognition Based On Deep Learning

Motion state recognition has always been a research focus in the field of state recognition.It has been widely used in virtual reality,health care,transportation and military training.At present,a large number of researches are based on traditional classification algorithms to classify motion states.Although the classification results are well,there are still many problems.For example,the data sources used are also relatively simple,and data acquisition and motion state recognition all require professional related equipment,which cannot collect and identify motion status data anytime and anywhere,limiting the scope of application.Based on the results of previous studies,a method of motio n state recognition based on deep learning is proposed by this paper.Firstly,the method uses the smartphone for data acquisition.And then uses different types of neural networks to construct a neural network model,where it combines deep learning techno logy and the Tensor Flow deep learning platform to complete the collection of motion data and recognition and finally this study uses the smartphone to realize the real-time recognition of motion state.The research contents of this paper mainly include the following aspects:(1)Human motion state recognition.Based on the published human motion state data set,this dissertation uses the Tensor Flow deep learning framework and deep learning technology to design classic neural networks,convolutional neural n etworks,long short-term memory neural networks(LSTM neurral networks)and bidirectional LSTM neural networks.The neural network completes feature extraction and classification of human motion state.By comparing the complexity of learning algorithms,structure of neural networks,computing performance of smartphone,and practical application of neural network models,which refers to four neural networks models,a neural network model suitable for motion in a smartphone is selected.(2)Identification of human traffic state.The continuous enhancement and popularization of the smartphone functions have reduced the requirements for data collection.This study uses sensors of the smartphone to collect human traffic stat e data,and denoises,segments,and organizes the collected data to produce corresponding training data sets and test data sets.Based on the recognition of human motion state,different types of neural units using deep-circulation neural networks are used to construct different neural network structures,where the parameters and the hidden layer unit data in the network are constantly adjusted during training,to complete the learning and classification of human traffic status data.According to the effects of training model recognition in the test data set,the final neural network structure is determined.(3)Model transplant and practical application.Combining Tensor Flow technology and Android technology,the trained neural network mod el is transplanted to the smartphone,and the data acquisition and real-time recognition of human motion state and human traffic state are completed through the mobile phone APP.