The Research On The Key Technologies Of Fall Detection

With the rapid development of economy and science,the aging of the population is becoming much more serious and has been a prominent problem.Meanwhile,the problem of the health and safety in the elderly has also aroused widely attention.Intelligent pension has become a major demand,since fall injury is a serious threat for the health of the elderly.On the one hand,effective and accurate fall detection can provide the elderly with timely treatment after falling,improve the quality of emergency,and avoid greater damage;simultaneously,it can decrease the occupation of medical resources and relieve the families' economic burdens.On the other hand,it can provide security assurance for the elderly,especially the empty nests,lighten their psychological burden,and enhance the quality of their later lives.This paper has carried out research and analysis on the key technologies of fall detection in the intelligent monitoring system and established the fall information database to provide support with fall detection algorithm.The fall detection algorithm has been studied emphatically,and a two-layer fall detection algorithm based on weight discrimination has been proposed.Multi-classification study of various activities in daily life has been analyzed as well.The main contributions of this paper are illustrated as follows:Firstly,measurement of activities in daily life(ADL)and falls are completed and a fall database is built based on wearable sensor systems.The database includes the information of 11 ADL and 4 fall activities,which can provide a benchmark for fall detection algorithm.Secondly,feature analysis is performed versatilely based on the established fall database.Appropriate features and the corresponding thresholds are selected,and the multi-threshold fall detection algorithm is designed and implemented.Experiment results verified the efficiency and feasibility of the proposed algorithm.Thirdly,the performance of the fall detection algorithms based on machine learning is compared and analyzed by combining the different classification methods and feature extraction methods.On this basis,classification functions of the artificial neural network(ANN)and the support vector machine(SVM)are improved by the.implement of AdaBoost.Additionally,the fall detection method based on ANN-AdaBoost and SVM-AdaBoost is proposed and the classification accuracy is improved.Additionally,a threshold detection algorithm based on weight discrimination is proposed since the above algorithm has been realized,which can reduce computation time and improve the recognition speed in the case of high accuracy.The performance of three kinds of fall detection algorithms is compared and analyzed as well.Furthermore,unlike above two-class classification problem,the research on multi-classification problem of various activities is introduced,and the multi-classification strategy is shown.The method of multi-class based on decision tree can identify each activity.The deep learning network model is explored.Also,the convolution neural network framework is proposed to recognize all kinds of activities.