Design And Implementation Of Fall Detection Application On Android

With the aged tendency of population, healthcare of the elderly has become more and more important in society. According to statistics, the elderly over 60 years old in China has more than 2 billion, accounting for the total population of 14.9%, and is expected to reach 400 million in 2024, and 4.8 billion in 2049. Due to the flux of young people in modern society, the elderly live separately without accompanying of their children, which may threaten the health of old people, once some unexpected things happen. Especially, for older people, falling is one of the leading causes of unintentional injuries. Therefore, timely detection of falling and rescuing the elderly quickly is a fundamental concern of our society.Nowadays, smart phones equipped with many kinds of sensors, such as acceleration sensor, magnetic sensor, light sensor, distance sensor, GPS, etc, and these sensors make smart phones with the conditions for fall detection. In addition, due to the communication function of the mobile phone, the rapid and effective rescue call can be carried out immediately after fall detection. In comparison with the invasion of wearable devices, environmental restriction of video detection and complexity of image processing, smartphone based fall detection has become the foucus of research in the industry.Based on the above research background, the application of fall detection based on Android mobile terminal is designed and implemented in this thesis. Users carry smart phones equipped with the application, and when a fall occurs the application will get the location, and then send a message automatically for help. Specifically, this research mainly includes the following four aspects:1. Fall analysis and model building. Fall is the vertical change of the body that is not controlled. Data changes of three-dimensional acceleration sensor during fall are analyzed in this thesis. Build a human body model, corresponding to the direction of the three dimensional accelerometer. When the body falls in any direction, the acceleration value of the direction can be decomposed into the component value corresponding to the direction of the three coordinate axes, and then the data acquisition is carried out by the acceleration sensor.2. Data acquisition and preprocessing. In order to obtain training data, we collect fall data for various situations with our developed software, including fall forward, fall backward, fall to the left side, fall to the right side, and fall under different height and weight, etc. Considerting that the original data collected intrinsically include noise and error, they are preprocessed by weighted smoothing and simple moving average method. In the case of obtaining the same smoothing result, the complexity of the former is O(1), and the latter is O(N), so this thesis chooses the weighted smoothing as the preprocessing method of the data.3. Fall detection algorithm analysis. The common activity recognition and classification algorithms are summarized such as Decision tree, KNN, Naive Bayesian, SVM, etc. The performance of the classification method is analyzed by Weka machine learning platform. Finally, the sensitivity, specificity and accuracy of SVM are 72.37%, 98.82% and 96.54% respectively, which can effectively detect the fall.4. Implemention of the fall detection system. Finally, this thesis designed and implemented a fall detection application based on Android phone, which can recognize the human activitiy in real-time and accurately.When fall occurs, the application can send a message containing the location information to the designated ambulance personnel, in order to get outside help.