Design And Implementation Of Fall Detection And Alarm System Based On Machine Vision

With the acceleration of the global aging process,the health and safety issues of the elderly gradually come into people’s vision,and have been widely concerned.Fall is one of the main threats to the health of the elderly,and it will cause serious damage to the elderly.At the same time,the elderly are often unable to recover after falling.If the elderly can get help or treatment in time after falling,the damage caused in the later period will be reduced a lot.Therefore,it is very important to detect the occurrence of falls timely and accurately.However,in the face of complex real-time environment,most fall detection methods based on machine vision are not ideal in real-time and accuracy.Based on the above background,this paper proposes a fall detection algorithm based on machine vision,which can detect the occurrence of fall behavior in a complex environment in real time.Based on this algorithm,a fall detection system is built,and the management of human information and alarm information is realized by combining Internet technology.In this paper,the main work is as follows:First of all,the overall design of the fall detection system,according to the system needs analysis,the whole system is divided into two parts: the fall detection software at PC end and the web information management system.The function of each part of the system is divided to determine the technical selection needed for the development of each part of the system.The algorithm of human body attitude estimation is studied,and openpose algorithm is used to detect the key points of human body.In order to improve the detection efficiency and accuracy of openpose algorithm,this paper improves openpose algorithm from two aspects: first,use lightweight network mobilenet instead of VGG network as the backbone network of the algorithm,and remove some fine detection network structure,further reduce the network model parameters and improve the detection speed;in the connection of key points,increase the key points Number of connections between,and rematch missing keys.Experiments show that the improved openpose algorithm has better detection accuracy for the human key points that are easy to be occluded,and it can detect the human key points in the surveillance video in real time.A fall detection algorithm based on multi feature fusion is designed and implemented.Firstly,by analyzing the differences between different behaviors,the key points of human body are detected to extract a variety of motion and morphological features.The extracted features include centroid motion feature,body angle change feature and star skeleton feature.Then we use relief algorithm to select the extracted features,and train the classifier according to the weight sorting.In the selection of classifiers,the random forest algorithm based on the integrated learning model is used.The experimental results show that the average sensitivity of the algorithm is 89.6% for fall detection and 89.4% for daily behavior detection.The algorithm in this paper can effectively detect the occurrence of falls,and it also has a good effect for similar falls.Finally,according to the previous research results,and combined with the actual needs.The fall detection alarm system is designed and implemented,which is composed of PC fall detection software based on Py Qt framework and Web information management system based on Django framework.At the same time,the database is designed and built to manage personnel information and alarm information.After functional test,the system can meet the needs of fall detection alarm and data management display,and it has high practical application value.