Research On Human Motion Posture Recognition Based On Mobile Phone Sensor

With the continuous maturity of sensor technology,sensors are gradually developing towards miniaturization,digitalization,intelligence and multi-function,which makes the human body daily motion posture recognition technology based on sensor signals get rapid development.Compared to appear earlier human movement behavior recognition based on video technology,using sensors as the source of data for research with data acquisition convenient,low power consumption,strong anti-interference ability and comfort good advantages,such as the early stage of the study into small and without violation of user privacy,therefore in the sports health,health care,human-computer interaction,etc,has the extremely widespread application prospect.As an indispensable communication device in people's daily life in modern society,smartphones' increasing computing and storage capabilities,as well as the sensory capabilities brought by rich sensors.These powerful functions have attracted the attention of many researchers to using smartphones as a platform.The daily movement behavior recognition of the human body has also become a new research hotspot.In this paper,the built-in sensor of the smartphone is used to obtain the acceleration information generated by the user's motion and the gravitational acceleration at the same time.The basic process of human motion posture recognition is studied according to data acquisition,data preprocessing,feature extraction and feature selection,classification model training and evaluation.The main contents of this paper are as follows:(1)Research and analyze the development and current situation of human motion posture recognition technology,and determine the objects to be identified in this study,including five basic activity states : walking,running,going upstairs,going downstairs,static,and four transition actions: standup,squat down,standing-Sitting and sitting-stand ing.For the nine kinds of sports to be identified,the sports data collection APP was designed on the Android platform,and the experimenters were invited to participate in the collection of experimental data to establish a human motion data set.(2)The user may change the direction of the sensor axis when using the mobile phone for motion data acquisition.The acceleration horizontal component and the vertical component are used to characterize the motion behavior to eliminate the influence of the sensor coordinate axis change on the experiment.(3)This paper proposes to use the trend test of the acceleration horizontal component combined with the DTW algorithm to detect the transition action in the segmentation sequence,avoiding the misclassification caused by the sampling window not including the complete transition action in the recognition process.(4)The time domain features of the horizontal and vertical components are separately extracted to form the feature vector,and the random forest classifier is used to classify and identify the nine behavioral states.Experiments show that the transition behavior detection and segmentation algorithm proposed in this paper can effectively segment the trans ition motion contained in the acceleration sequence,and the method in this paper is used to process the acceleration data.Finally,the average recognition rate of nine kinds of motion behaviors reaches 97.26%.Compared with some related researches in recent years,the recognition effect has been significantly improved.