Research On The Fall Detection Algorithm Based On Accelerometer

The advent of aging society has brought about ever increasing attention paid to elderly people's health.As the elderly falls may lead to serious consequences,people have begun to research the fall detection equipment.Commercially available fall detection equipment fails to provide accurate assessment,which contributes to more research in fall detection.This paper researches the accelerometer–based fall detection algorithm to improve the fall detection performance.After researching algorithms that based on time domain features and wavelet features,an algorithm based on nonlinear features is proposed.The main contents and research results as below:First,the acceleration data is preprocessed through threshold-based method,thus minimizing the difficulty of subsequent fall detection effectively.Second,KPCA(kernel principal component analysis)algorithm is used to extract nonlinear features of tri-axis acceleration data separately.And the improved KPCA algorithm is proposed to extract the contact information between the tri-axis acceleration data.Third,nonlinear extracted features are classified by different classifiers to choose the optimum classifier.In order to further improve the recognition rate,the SVM(Support vector Machine,SVM)and the K-Nearest Neighbor(KNN)algorithm are combined to vote in some difficult-to-distinguish samples according to the output result of Probabilistic SVM.Finally,characteristics of the after-fall state are used for validation to further improve the accuracy of detection.The proposed algorithm has been verified on the UCI Machine Learning database,the detection accuracy rate reaches 96.17%,and the recognition rate of fall reaches 96.89%.In local experiment,the detection accuracy rate and the recognition rate is 96.36% and 97.50% respectively.