Research On Fall Detection Methord Based On Support Verctor Machine And The System Design

With the advancement of medical technology and the improvement of living standard,China's population aging is becoming more and more serious,and the proportion of the elderly living alone is also rising gradually.If the elderly were not found in time to when they fall,and missed the best treatment period,which not only can cause serious physical and mental damage,and may even lead to death,Therefore,the study of real-time fall detection has important social significance in improving the elderly's ability to live alone and their health.Now,the common falls detection technology at home and abroad can be divided into fall detection based on environmental information;videobased fall detection and portable fall detection.The fall detection based on environmental information and video-based fall detection can display the human body's posture clearly and have high accuracy,but they expose too much user's privacy,and the monitoring range also has some limitations;The portable fall detection is cheap and simple to use,and it can detect falls at anytime and anywhere detection,which make it be used widely.The most common portable fall detection device is inertial sensor fall detection system based on the threshold algorithm,but its accuracy is low,the false rate and false negative rate is higher,which leads poor user experience.In order to solving the problems above,this paper proposed a new fall detection algorithm to improve the accuracy and reduce the false positive rate and false negative rate of the fall detection system based on inertial sensor.The main contents of the paper include:Part one: hardware design.The real-time fall detection system in this paper used the embedded processor STM32F103 as microprocessor,used three-axis accelerometer and gyro sensors to realize the data collection,and used SD card and GSM module to complete the data storage and remote distress respectively.The system powered by rechargeable lithium battery,which can work more than 24 hours sustainably.The overall size and power consumption of the system meet the portable requirements.Part two: eigenvalue extraction.In order to better distinguish fall from daily motion accurately,we extracted five characteristic parameters based on the original data including the acceleration SV,the dynamic acceleration SVD,the vertical acceleration component DVA,the acceleration change K and the attitude change M.Final results showed that the five eigenvalues above improved the classification accuracy greatly.Part three: algorithm design.In this paper,we used the MATLAB tool to complete data analysis and algorithm simulation.Firstly,we used SVM classifier to mark the suspected fall behavior.Then used the particle swarm optimization(PSO)algorithm to complete the optimization of penalty factor C and radial basis parameter g.Finally transplant the best parameters into the offline system.Experimental results showed that the algorithm can distinguish fall from daily behavior well.The experimental results showed that the accuracy,false positive rate and false negative rate of the fall detection system based on PSO-SVM algorithm are 97.67%,4.0% and 0.67% respectively.In comparison with the traditional threshold based system,the performance is greatly improved,which strengthened the system's application in the elderly falls detection fields greatly.