Research And Application Of Human State Recognition Technology In Home Enveronment

Recently,China has gradually stepped into an aged society.As its economy grows,the number of migrant workers has increased,the number of only children has increased,and the number of elderly people living alone has also increased.In the course of soliloquy,the physical condition of the elderly is deteriorating,injuries are neglected,and other reasons make them prone to some accidents.Therefore,the importance of the research on the elderly care system has become increasingly prominent,while the human posture recognition technology is the core content of the elderly care system.In this paper,the method of machine vision is used to recognize the pose of the human body.At present,there are not many gesture recognition methods based on machine vision and artificial intelligence neural network.The research in this paper can effectively combine the two methods to improve the recognition accuracy.The Xtion infrared camera used in this article is a machine vision category that quickly acquires the data it needs and protects the elderly from the equipment during identification.Human pose recognition technology is affected by many factors,including light,distance,and camera distortion.In this paper,we use the infrared camera Xtion Pro Live to study the posture of the human body.It is affected by the light and has no interference with the distance,so it can be detected in the indoor environment.Xtion Pro Live infrared camera through the point cloud array can get the depth map in space.Depth map is used to represent the depth of field in grayscale image,each pixel represents the space object at this point to the camera distance,in order to obtain a three-dimensional coordinate image.The collected images will be noisy,using Gaussian filter to filter the image,the image to reduce noise,more clearly.After obtaining the depth map,the background of the image is modeled,the human target information we need is separated from the image,and the feature of the foreground target is extracted as the basis of the classification.This article should identify the human body standing,lying,sitting,falling and other actions,the use of human body center of gravity three-dimensional coordinate information and head orientation information as a feature value to be classified.Multiple sets of data were extracted for each status sample as samples for training.This article uses BP neural network algorithm in artificial neural network to do classification and recognition.BP neural network algor ithm has the advantages of accurate classification and high accuracy of nonlinear classification.In the process of training,the gradient descent algorithm is adopted to adjust the parameters and the recognition of the dynamic fall action is added.After the completion of the algorithm design experiments in the indoor environment,the experimental results show that: the stability and accuracy of the system are relatively high.Finally,the algorithm system is applied to the development board and performance analysis.