Fall Detection System Design Based On Multi-sensor Feature Fusion And SVM

With the aggravation of aging of population,the design of the fall detection system for the elderly living alone has important research value and social significance.With the wide application of video monitoring in the society and the miniaturization of sensor devices,using multisensor data to detect falls has become a new method for the safety monitoring of elderly people living aloneMainstream methods about fall detection include video-based,wearable and environment-based sensors,all of which have their own advantages and disadvantages.On the basis of existing fall detection methods,this paper designs a fall detection system based on multi-sensor feature fusion and SVM(support vector machine).The system mainly performs the following tasks1.A collection device based on accelerometer and camera is designed.The device collects a large amount of human motion data through different sensors,and mark the current human falling state as the data set,so as to solve the problem that the common falling detection database does not contain multi-sensor data2.Background subtraction and other methods were used to extract the foreground contour of human body from the collected video data.Using the acceleration data and the human body contours,the sensitive features of falls are extracted.The problem of data matching in multi-sensor fusion can be solved,and feature vectors are constructed to provide samples for SVM algorithm.By using different constructing methods,the performance of each feature can be analyzed,and the eigenvectors with the best performance are selected3.Use SVM algorithm to detect falls.Part of the data set was used to train SVM to obtain the training model,and then the training model was used to classify the rest of the data set and detect falling of human body.SVM algorithm is used to test the detection method based on a single sensor and the system method introduced in this paper,and to analyze their advantages and disadvantages in falling detectionIn terms of hardware design,this system considers the actual use,and designs for data collection,reception,fall detection and other steps,effectively ensuring the real-time and stability of the system.At present,most fall detection methods are based on single sensor to extract features and use threshold judgment or machine learning to detect falls.This system uses a variety of sensors to collect data,and uses the method of multi-sensor fusion,combined with SVM algorithm for fall detection.Compared with the current method,the system method significantly improves the amount of information acquisition.The system uses fusion features to make efficient use of each sensor information and improve the efficiency and accuracy of SVM algorithm.Finally,the system has obvious advantages in sensitivity,false alarm rate,especially in classifying falling behavior and falling behavior.