Application And Research Of Behavior Pattern Recognition Algorithm Based On Smartphone Platform

With the rapid development of microelectronic information system technology,different sensors are miniaturized into various wearable devices,as well as portable smart devices.By mining the data of the built-in sensors in these devices,can analyze user's behavior,which provides a basis for researching the recognition algorithm of human activity based on the smartphone platform,thereby developing personalized and intelligent services such as patient rehabilitation training,athlete posture correction and daily human behavior monitoring,with broad market prospects and economic benefits.Human behavior recognition algorithms are developed commonly based on machine learning algorithm technology,this kind of algorithm can solve the problem of low recognition rate in the field of behavior recognition due to differences in individual exercise rate,exercise posture,and individual height and weight.This paper proposes a human behavior recognition convolutional neural network algorithm with machine learning ability,which can recognize six kinds of human daily behaviors,including walking,jogging,sitting,standing,going upstairs and going downstairs.The key aspects of the development include the acquisition of human behavior dataset,the preprocessing of dataset,the training process and the performance testing of neural network model,and the optimization of model structure and related parameters.In terms of algorithm performance evaluation,this paper uses LIBSVM platform to build a behavior recognition system based on SVM algorithm.The recognition rates of CNN and SVM algorithm systems are obtained by running the WISDM standard classifier data set.The results show that the accuracy of the recognition system based on CNN algorithm can reach more than 90%,which is obviously better than the SVM algorithm recognition system with accuracy of about 80%.In order to verify and apply the results of the algorithm,we develop a behavior recognition APP based on the Android smartphone platform,which can track,record and analyze the daily behavior of the phone carrier in real time,and according to the recognition results,APP provides many health-assisted functions such as sedentary reminders and exercise statistics.In the software running test,the statistical result of the CNN algorithm for the above six daily behavior is 88.65%,showing excellent application value.The relevant content of the thesis has applied for intellectual property achievements such as algorithm patents and copyrights.