Research On Fall Detection Based On Wearable Sensing

China has entered an aging society,and the elderly are prone to falls.Falling not only endangers the health of the elderly,but also makes them fearful of activities,thus making their physical conditions go from bad to worse.The fall detection equipment can detect the fall of the elderly in time,so as to send the fall information to the medical team of the elderly,and minimize the continuous injury of the fall to the elderly.Therefore,fall detection technology is of great significance to the health of the elderly.This article uses wearable sensing technology to detect falls,and is committed to improving the accuracy of fall detection and reducing the false alarm rate of fall detection.Mainly study the motion signals and stress signals of the elderly,and find ways to improve the accuracy of fall detection and reduce the false alarm rate of fall detection through experiments.The research results are expected to help the lives of the elderly to be safer and improve the daily lives of the elderly to a certain extent.The specific research content of this article is as follows:(1)Research on wearable sensing technology: This article uses common sensors to collect fall information.Sensors such as accelerometers,gyroscopes and barometers are placed on the waist to collect the motion signals of the elderly’s torso;pressure sensors are placed in the shoes to collect Pressure signals on the feet of the elderly.In this paper,Kalman filtering method is used to process the motion signal to reduce the influence of noise on the output signal of the sensor.Complete the feature extraction of the motion and pressure signals,and use the random forest algorithm to select the features that contribute greatly to the fall detection.(2)Fall detection algorithm based on multi-source information fusion: This paper confirms that a single state sensor cannot well recognize similar fall actions such as sitting,squatting,and lying down.Therefore,a fall detection system that integrates multi-source information is proposed.The attitude angle signal and the pressure signal source are fused,and the multi-state threshold method is used to filter out daily actions,that is,the fall threshold that meets the acceleration vector and the average value of the heel area pressure at the same time will be judged as a fall action.Extract 21 features such as mean value,sum vector magnitude,horizontal direction vector sum and root mean square and send them to the random forest algorithm for feature importance scoring,select features with a score greater than 0.1 as the input features of the extreme random tree,and use the extreme random tree do further fall detection.Experimental results prove that the system can recognize falls well in similar fall actions,with an accuracy rate of over 98.3%.(3)Fall detection software design based on machine learning: Software for fall detection is written in the host computer,including data collection,data visualization,fall detection and other functions.The software integrates six machine learning algorithms,namely KNN,SVM,decision tree,random forest,deep forest and extreme random tree.Through experiments,the accuracy and time required for fall detection of the six algorithms are compared.The results show that the use of extreme random trees for fall detection and recognition has higher accuracy and less time.