Research On Intelligent Fall Monitoring System Based On Deep Learning And WebRTC

As China's population aging continues to increase and the growing pressure on social life,the proportion of empty-nesters continues to rise.The elderly living alone cause a lot of social problems,especially the health problems of the elderly,in which the elderly accidental fall injury is one of the main reasons affecting the health of the elderly.The elderly did not fall after the treatment will increase the secondary injury,and even lead to accidental death of the elderly.According to the relevant statistics,more than 50% of the elderly fall at home,if the elderly can fall in real-time monitoring behavior,when the elderly fell after the behavior can be accurately identified,and to send their reminder to the guardian,so that the elderly get timely relief Greatly reduce the damage to the health of the elderly.Based on this,this paper studies the intelligent fall monitoring system based on depth learning and WebRTC.The main work is as follows:After analyzing the functional requirements of the intelligent fall monitoring system based on the depth learning and WebRTC,the technical scheme of the system is put forward.The program uses the depth of learning technology to achieve the elderly people fall behavior of intelligent identification,based on Web RTC video transmission architecture to achieve remote video transmission.A study on fall detection based on depth learning is carried out.Firstly,the fall detection method based on video frame and VGGNet-16 convolution neural network model is proposed and simulated.The data of Le2 i,SDU and UCF-101 open source video datasets are horizontally inverted,contrast and brightness adjustment,After training,the training dataset and test dataset of neural network are constructed,and the method is trained and tested.The experimental results show that the method is strongly dependent on the training scene.Secondly,aiming at the above problems,a fall detection method based on double stream convolution neural network is proposed.The method comprises the following steps of: using the background subtraction method to detect the moving target in the video,marking the moving object in the video frame into the 3D-CNN model for the fall recognition;the other way adopts the optical flow method to extract the optical flow chart of the video,The optical flow graph is input into the VGGNet-16 model for the fall recognition.Finally,the results of the fall recognition of the two models are linearly weighted.The experimental results show that the fall recognition rate of the fall recognition method based on the double flow convolution neural network is 96%,which is 51% higher than that of the video recognition method and the VGGNet-16 convolution neural network model,which is higher than that of the moving object And 3D-CNN fall recognition method increased by 4%,compared with the optical flow chart and VGGNet-16 method to improve the fall of 3%,and a good generalization ability.A study on remote video monitoring based on WebRTC is carried out.A remote video monitoring scheme based on WebRTC is put forward,and the signaling server and the server for the video transmission are built.The video acquisition terminal and the remote video monitoring terminal are realized based on WebRTC.Through the network experimental environment containing NAT on the system test,the experimental results show that: the implementation of the server,network server in the normal work,the client of video transmission function,and can pass through the firewall and NAT restrict the realization of video transmission P2P.