Research And Application Of Human Activity Recognition Technology Based On Android Platform

With the rapid development in today’s society, people pay more attention to their health while busy with work. As a new research direction, the human activity recognition based on sensor has greater social benefits while monitoring the daily activities of human body in real time. On the other hand, China’s population aging problem is increasingly outstanding, and it offer occurs the event which caused by fall that seriously threatens the old people’s physical and mental health. Therefore, it is very significant to achieve a system which can do effective detection when people is tumbling and provide timely help after people’s tumble. In recent years, with the rapid development of micro-electronics technology and the mobile Internet, most of the smart phones are embedded with a lot of sensors, which makes the related research above become more convenient.On the basis of the Android embedded single acceleration sensors, this article does some research and make some improvement about the human body movement recognition mechanism and tumble detection algorithm, the main contents are as follows:Firstly, based on analyzing and summarizing the existing mechanism of the human body activity identification, this article does the research on human body recognition model based on the Android platform, including:at first collecting real-time motion acceleration data by programming, and doing some preprocessing operation to the original acceleration data such as adding window, smoothing the original acceleration data and separating the linear acceleration and so on; then analyzing the acceleration signal and extracting ten kinds of acceleration characteristic values from time and frequency domain separately, and doing the dimension reduction with the principal component analysis; Finally, recognizing the daily behaviors such as walking, jogging,climbing stairs, descending stairs and so on by the classification algorithm effectively. At the same time, this article verifies C4.5 decision tree, Naive Bayes, LogitBoost and SVM algorithms based on other researchers’data sets, and compares with their experimental results.Secondly, by analyzing the body posture of human when they are tumbling, this article puts forward a multiple threshold tumble detection algorithm based on the Android platform. The algorithm firstly does three-step moving average smoothing preprocessing to the gathered movement data, and then extracts the body’s front and back pitching angle, left and right sides side-slip angle and maximum acceleration, acceleration value after a tumble, and five thresholds of continuous rest time in order to detect the behavior of fall.The experimental results show that the proposed tumble detection algorithm can effectively identify tumble behavior from daily activities and has certain practical value.Thirdly, using the proposed tumble detection algorithm based on the Android platform implements a human body tumble detection system.The system does real-time monitoring on human behavior. Once detects a tumble, it’ll immediately issue a buzzer alarm sound and send messages to seek help from the specified contact person to make sure the tumbling person get timely treatment. Thus it can be seen that this system has a certain application prospect.