Research On Human Activity Recognition Based On Multi-sensor Fusion Technology Of Smartphone

Human Activity Recognition(HAR)is an important research field in pervasive computing,which is widely used in human-computer interaction,medical diagnosis and abnormal behavior Recognition.Based on environment awareness,HAR's sensors(such as cameras and infrared sensors)are deployed in a limited environment and are only suitable for specific scenarios.HAR technology based on visual perception is limited in monitoring scope and easy to expose user privacy,which makes it difficult to apply and popularize the technology.HAR technology based on wearable awareness can operate in both indoor and outdoor environments,but deploying multiple sensors on the body is costly and can cause discomfort for the wearer.In recent years,smart phones are increasingly integrated into people's daily life and become a must-have for people's life.With the development of mechatronics technology,various kinds of high-precision sensors are becoming smaller and smaller,which makes smart phones integrate more and more sensors,such as 3-axis accelerometer,gyroscope,magnetometer,thermometer and barometer,while computing power and storage space are rapidly increasing.In this regard,this paper makes use of the powerful computing function and perception ability of smart phones to study HAR technology based on smart phone multi-sensor data fusion.The main work and achievements are as follows:(1)The motion model of human body based on 3-axis acceleration,angular velocity and magnetometer is established by cartesian coordinate system.The framework realizes the preprocessing and feature extraction of multi-modal human activity sensing data through data filtering,normalization and sliding window.(2)For the multi-mode data of human activities collected by smart phones,features were firstly extracted from the time domain and frequency domain,and features were selected by principal component analysis.Secondly,a better human activity recognition model was obtained by designing Stacking integration algorithm to train the collected multi-mode data.(3)The multi-sensor data fusion HAR system was designed,and the human body recognition data set was disclosed by Mobi Act for experimental test,and the experimental results were compared with other algorithms.Experiments have proved that the algorithm in this paper can accurately identify daily activities such as standing,walking,jogging,going upstairs,going downstairs,riding a bike,etc.,with an average system accuracy of 99.0%,average sensitivity and average specificity of 99.0% and99.8% respectively.Its recognition accuracy is higher than that of Adaboost,random forest,support vector machine,k-nn and other algorithms.In this paper,a motion recognition system of human body is designed andrealized by using the 3-axis accelerometer,gyroscope,magnetometer and barometer of smart phone.The system has the advantages of easy to use,high recognition accuracy and good application prospect.