Research On The Detection Algorithm Of Elderly Fall Behavior Based On Multi-feature Fusion

According to China's national conditions,the aging phenomenon will lead to more and more elderly people in the future.As one of the biggest threats to the health of the elderly,falls not only do serious harm to the mental health and physical health of the elderly,but also take the lives of the elderly.Therefore,it is of great practical significance to detect the fall of the elderly in real time with the help of intelligent technology and minimize the injury of the elderly.Through literature review,this field has two difficulties: first,most of the current research methods are based on computer vision or single sensor fall detection based on sensors,which can not overcome the impact of environment or forget the problems of wearing sensors.Second,most of the literatures are based on statistical characteristics such as mean and standard deviation.In the state of motion,the nonlinear dynamic characteristics that may exist in the process of falling are not considered.Aiming at the above problems,this paper designs a multi-modal fall detection system combining computer vision and sensors,introduces the idea of feature fusion,proposes a fall detection algorithm based on multiple features and fuzzy KNN,and completes the three states of fall,non fall and fall like detection experiments.The algorithm has achieved good detection results in the simulation and the actual system,and in one To some extent,it overcomes the problems of environment and forgetting to wear sensor,and provides new ideas for the future in this field.The main contents of this paper are as follows:1.A multimodal database for fall behavior detection is established in this paper.The secondary feature extraction algorithm based on the combination of statistics and entropy is proposed.The representation ability of entropy to fall data and its contribution to detection rate are studied.The necessity of using entropy feature and its relation with actual behavior state are analyzed by multi-scale composite fuzzy entropy curve.2.In this paper,fuzzy KNN is introduced to fall detection for the first time.The effects of each parameter of the fuzzy KNN classifier on the three behavior states of falling,falling like and non falling are studied,and the performances of different classifiers are compared.The experimental results show that the proposed feature combined with fuzzy KNN achieves the best recognition results,with 98.6% accuracy and 1.0% fall error rate.3.Taking the above algorithm research as the theoretical basis,this paper develops a real-time and scalable fall behavior detection system with C language based on opencv3.0 and vs2017 platform,which verifies the effectiveness andreal-time performance of this algorithm.Aiming at the real-time system,this paper gives a comprehensive test result,and the accuracy of the system's fall behavior detection and recognition reaches 97.5%.On the one hand,the algorithm proposed in this paper can effectively reduce the error recognition rate and solve the problem that the system can't alarm when the elderly fall;on the other hand,the algorithm proposed in this paper can effectively overcome some problems caused by occlusion and other environments,and solve the situation that the elderly fall when they can't be recognized by the camera head.It provides an effective solution to improve the health monitoring of the elderly.