Research And Application Of Human Activity Recognition Technology Based On Intelligent Mobile Terminal

With the increase of the ever accelerating pace of life and work pressure, a growing number of people attach importance to the healthy lifestyle. Monitoring of the human daily exercise condition has very important practical significance for the healthy way of life.Nowadays, with the rapid development of science and the continuous improvement of manufacturing technology, more and more researchers pay attention to the human activity recognition technology based on intelligent wearable devices. Today, smart phones not only have a strong computing power, but also integrated a variety of sensors. Based on this, this paper presents the research and application of human activity recognition technology based on smart phones.This thesis mainly use the smart phone which built-in acceleration sensor. Based on the analysis of the domestic and foreign classical methods of human activity recognition, this thesis studies and improves the human activity recognition method and the fall detection algorithm. And realizes the human activity recognition system based on the smart phone sensor.Firstly, the acceleration signal of human activity is collected. And the original acceleration data is preprocessed by adding window,smoothing and so on. Then, analyzing the acceleration signal and extracting acceleration characteristic values from time and frequency domain, and using the linear discriminant analysis to reduce the dimension of feature set. Finally, the six daily behaviors of walking, running, resting, cycling, upstairs and downstairs are classified by K neighborhood, C4.5 decision tree, SVM and random forest classifier. The experimental results show that the average recognition accuracy of SVM is the highest, reaching 94.06%.Secondly, this thesis proposes a hierarchical fall recognition algorithm based on SVM by analyzing the acceleration signal and the change of human posture in the process of human fall. The algorithm first extracts the features of the activity acceleration during the fall process, including the maximum acceleration, mean-crossing times, signal magnitude area and the signal energy. The SVM classifier identifies the suspected fall behavior, and then determines the suspected fall behavior whether the human body tilt angle reaches the set threshold, which can accurately identify the human body's fall behavior. The experimental results show that the fall detection algorithm proposed in this thesis has high sensitivity and specificity, which can effectively identify the fall behavior of the human body.Finally, on the basis of the activity recognition method and the fall recognition algorithm proposed in this thesis, the human activity recognition system based on Android smartphone platform is developed and realized. The system (1) can be real-time classification of daily behaviors of users, and through the Baidu map for user positioning and display of movement routes; (2) can be the user's fall behavior detection, once the user to identify the fall will be issued immediately with a buzzer alarm sound, and send help messages to the emergency contact, so that users can get timely assistance. The human body motion recognition system realized in this thesis has certain practical value.