Design Of A Multi Feature Fall Detection System Based On Improved Openpose Network

After entering the 21 st century,China’s economy is in the stage of rapid development,the medical environment is improving day by day,and the medical technology is progressing day by day.The following is the change of Chinese fertility concept.China’s birth rate is decreasing every year,the aging population is increasing,the resources of pension institutions are gradually scarce,and the pressure on this aspect is increasing day by day.In this case,home-based care will become an important way to solve the care of the elderly.According to relevant data,abnormal falls caused by sudden diseases are one of the important reasons for the death of the elderly living alone at home.Under the family monitoring environment,the abnormal fall detection of the elderly living alone at home has naturally become a hot research direction in the field of computer vision and image processing.This paper is a home-based elderly fall detection system based on a multi feature fusion algorithm designed based on the improved openpose network.It focuses on the application of the system in the scene of home-based elderly care service.By collecting the video of daily activities of the elderly living alone,the elderly fall behavior can be monitored and identified in real time.Once an abnormality occurs,Immediately send the location information and abnormal fall information of the elderly to the guardians and community medical staff at the same time,and take the initiative to carry out necessary treatment in time to reduce the injury caused by falls of the elderly.Due to the complex environment in the family scene,the number of elderly people living at home will be more and more large in the future,which requires higher accuracy of system recognition.Based on the analysis of the traditional convolutional neural network and the first generation openpose human skeleton information recognition network,this paper first improves the network architecture of openpose,and designs the multi feature fall discrimination basis.A large number of experiments show that the system meets the expected requirements.The specific work is as follows:(1)This paper introduces the research background and significance of fall detection for the elderly,the research status at home and abroad,makes a systematic research and investigation on the relevant knowledge of deep learning,and introduces the basic theory and network structure of several classic convolutional neural networks.This paper studies the original openpose skeleton recognition method,puts forward its shortcomings in fall recognition,and improves the network structure based on it.(2)On the basis of openpose human skeleton information recognition network,this paper proposes to replace its original backbone network vgg-19 with the type of deep separable convolution neural network,and selects mobilenet.However,mobilenet itself is a lightweight network.The experiment shows that the accuracy of direct replacement will be reduced.Therefore,two improvements are made on the basis of mobilenet: first,the residual structure is added;Secondly,the activation function is improved.(3)For the new fall detection network proposed in this paper,a multi feature fusion algorithm is designed.The four evaluation features are: the Euclidean distance from the normalized coordinate set to the reference point;Length width ratio of human external matrix;The included angle of specific parts;Descent speed of key points of human body.With the evaluation criteria under a variety of features,it can effectively avoid the misjudgment and accuracy reduction caused by a single feature.