Research And Application Of Fall Detection Algorithm Based On Machine Learning

With the improvement of people's living standards,the aging of the population is more and more serious.Falling is a very serious problem for the elderly.Real time monitoring whether the elderly fall is of great significance to reduce the injury caused by falling.Because the traditional fall detection method based on threshold characteristics can not fully describe the fall behavior,which leads to the low accuracy of the fall detection algorithm,and there are a lot of false positives,so it is necessary to design an efficient fall detection algorithm.Aiming at the problem of high error rate of traditional threshold detection method,this thesis studies the machine learning based fall detection algorithm to optimize.The model instability caused by the traditional manual extraction of threshold characteristics of fall can be trained by machine learning algorithm,combined with a variety of dominant and recessive features to detect the perception of fall,which can solve the problem of misjudgment and false alarm.However,there are many algorithms in the field of machine learning,and there is not a widely recognized algorithm for fall detection.Therefore,based on the research of the algorithm for fall detection based on machine learning,this thesis designs a fall detection algorithm based on LM-BP neural network,and uses three data sets to evaluate the algorithm in terms of accuracy and time complexity.Compared with other machine learning algorithms,the falling detection algorithm based on LM-BP neural network takes the time complexity into account on the basis of ensuring the accuracy,which is better than other falling detection algorithms based on machine learning in general.In order to test the feasibility of the algorithm,this thesis designs a wearable waist tag device,which can be connected to the Internet through GPRS and equipped with motion sensors,can detect the acceleration and angular velocity signals of the wearer in real time,and upload the signals to the cloud server,which calls the fall detection algorithm,once the wearer is perceived to fall,the ECS will send out an alarm message and locate the position of the fallen person through GPS.The main contents of this thesis are as follows:(1)This thesis analyzes human activity behavior,divides human activity into fall behavior and daily activity,and puts the sensor on the waist.This thesis also carried out data collection experiments,established a data set of falls and daily behaviors,and used Mobi Act and Sis Fall two public data sets to evaluate algorithms,it also introduces the traditional algorithm of fall detection,and further explains the disadvantages of the traditional algorithm combined with the data.After that,the data in the data set is preprocessed and feature extracted for data input of the BP neural network,and then the neural network training is completed using the LM algorithm.Experiments show that the accuracy of the algorithm based on LM-BP neural network can reach more than 99%.Compared with other machine learning algorithms,the algorithm based on LMBP neural network takes both accuracy and time complexity into account,which has good application value.(2)This thesis designs a wearable device based on MTK platform,which has the functions of GPRS Internet access,acceleration and angular velocity acquisition,GPS positioning,etc.This thesis completes the hardware and software design of the device,and successfully applies it to fall detection.(3)Finally,the wearable device and the fall detection algorithm are tested jointly.The experiment shows that the fall detection device and algorithm in this thesis achieve the function of automatic fall detection and alarm,and achieve the expected goal.