Research And Implementation Of Fall Detection Method Based On Smartphone

Accidental fall is an incident that seriously jeopardizes the health of the elderly.Compared with the traditional active alarm equipment for the elderly,the smart phone is equipped with the functions of fall detection,automatic alarm,active positioning,etc.The elderly may lose the ability to seek help after falling,especially in outdoor situations.And the advantages of the smart phone are more obvious.The core of smartphone fall detection research is to implement fall detection method.The research on human body fall detection based on smart phones is still in a relatively early stage at present.Many problems that need to be solved still prevent the use of fall detection algorithms in real life.Including how to design the best fall detection algorithm for the wearing position of the smartphone;research on the robustness and stability of the algorithm;the establishment of multi-wear position,multi-sensor database;performance and power consumption of the fall detection algorithm on the smartphone etc.This paper will focus on these difficulties and conducts a series of research on the fall detection method with smart phones.The main work includes:1.Proposed human accidental fall detection method based on the wearing position of a smart phone.Firstly,the rotating mode component and attitude angle fusion algorithm(FRMAA,Fusion of Rotation Mode and Attitude Angle)is proposed to distinguish the wearing position of the smartphone;then a Time Series Fall Detection Analysis(TSFDA)is proposed,which utilizes the algorithm.This method uses different feature subsets to construct models in different wearing positions to optimize the effect of each position drop detection algorithm.2.This paper constructs a basic data set that supports human fall detection and identification.The data set is collected by multiple smart phones simultaneously in different positions of the human body,including fall positions and ADL(Afficiencies of daily living)data at various positions such as wrist,waist and pocket.The data set was named as the Zhejiang University multi-position position human fall action database.3.For the noise problem caused by the error of Fall and ADL sample mark in the current open source fall detection data set,by adjusting the regularization parameters,the TSFDA algorithm can use the incremental data to improve the performance of the learning machine.In this part,we first explain theoretically the reason why the support vector machine(SVM)is robust to class noise,and use the data in the ZJU-MWPFA database to verify that the TSFDA algorithm proposed in this paper has a relatively high robustness.4.Development and implementation of a human body fall detection prototype system based on the Android platform.In the Android operating system,human fall detection prototype system based on the FRMAA and TSFDA algorithms was developed.The prototype system was tested and analyzed in different brands of Android phones.It was found that the system can be stably operated in various models of Android phones.Under each wearing position,the fall detection and recognition effect is basically consistent with the theoretical results.