Research Of Wearable Fall Detection System Based On MEMS Inertial Sensor

At present,China's population structure has entered an aging stage with the proportion of elderly people increasing.Fall accident of elders has became a serious problem that has plagued families and society.Elder people could suffer from injuries and many sequelae because of fal s witch would cause serious damage to both physical and mental health and leading to serious family burdens.In addition to prevention,needs timely treatment after fal s is necessary to reduce old people's fal ing injuries.It has become a worldwide hot topic to detect and predict fall by researching the fall process of the elderly through scientific methods.Regarding the issue above,In this paper,a Wearable fall detection system based on MEMS inertial sensor was proposed,which could alert elderly before they fal,and sends the warning information to the monitoring platform.the main jobs are listed as follows:Firstly,This paper explains the trend of fall injury of the elderly,and studies the existing fall detection system and fall detection algorithm with analyzing its advantages and disadvantages.The system use a nine-axis acceleration sensor as the acquisition core for human kinematics data.Secondly,I,in this paper,divide human behavior into normal behavior and fall behavior according to the type of behavior,and divide human behavior into ordinary behavior and violent behavior by the change of kinematic data.We establish a Cartesian coordinate system and solute attitude by quaternion combines first-order complementary filtering algorithm.Thirdly,Software and hardware of the system has been designed in this paper,The hardware of the system was designed by connecting Microcontroller module,MEMS inertial sensors module,GPS/GPRS module,Bluetooth module,Power module and alarm module.At the same time we designed the circuit diagram and PCB board and make a physical sample,also designed and test acquisition kinematic data,fall detection and alarm procedures.Fourthly,after analyzing the kinematic data for different behaviors,we extracte the combined acceleration and the combined posture angle as classification features,and using the threshold algorithm to determine the fall behavior.Finally,we experimental test Multilevel threshold detect algorithm and SVM-based detect algorithm respectively,witch shows the sensitivity,specificity,and accuracy of the Multilevel threshold detect algorithm are 91%,92.33%,92.27% respectively,meanwhile,the sensitivity,specificity,and accuracy of the SVM-based detect algorithm are 99%,96.67%,97.73% respectively.We draw the conclusion: the result of SVM-based detect algorithm is better than Multilevel threshold detect algorithm's.