Research Of Real-time Fall Detection Algorithm And System Based On Inertial Sensors

With the rapid development of modern society,the problem of population aging has become increasingly prominent.The new “4+2+1” family structure makes the traditional pension mode no longer applicable.Therefore,the problem of elderly care has attracted more and more attention.Falls are one of the leading causes of physical injuries or death to the elderly [1].Previous research shows that a third part of the elderly over 65 suffer at least one fall per year [2] while 4%-15% of falls result in severe injuries,and 23%-40% of injury-related deaths of the elderly are due to falls [3].Most falls do not immediately cause serious harm.Sometimes it is caused by the lack of timely rescue after the fall.Therefore,although it is difficult to avoid the occurrence of falls,the earlier the fall is alarmed,the lower is the rate of severe injuries or death.In this paper,we fully investigated the fall detection problem,and we designed a fall detection system based on inertial sensors.The main content of the paper are as follows:1.We fully investigated the existing fall detection algorithms and systems,analyzed the advantages and disadvantages of different methods,and designed a fall detection system based on inertial sensors and intelligent terminal devices.2.Since manually designed features usually limit the accuracy and time delay of the algorithm,a fall detection algorithm based on recurrent neural network is proposed,which can automatically extract features from raw data.The algorithm was tested on two public data sets to verify its feasibility,accuracy and time delay.3.In order to further improve the generalization ability of the fall detection algorithm,we introduced frequency information to the recurrent neural network,so that the algorithm can extracts the features of the time series sensor data in both the time domain and the frequency domain.We also verified the generalization ability of the algorithm with two public data sets.4.In order to realize an end-to-end,edge computing fall detection system on intelligent terminal devices,we designed and developed a real-time fall detection APP based on Android smartphone and Tensor Flow Lite.