Research And Implementation On Fall Action Analysis Technology In Household Scenarios

With the development of computer network technology and the gradual improvement of intelligent terminal equipment,it is possible to create a smart,safe and convenient home life.Falling is a dangerous action that often happens in household scenarios.How to quickly and accurately detect fall action has become a hot research area in society.In recent years,computer vision and deep learning technologies have developed rapidly,providing a new way to solve this problem.Based on this popular field,a extraction algorithm of human silhouettes based on the optimized hybrid Gaussian model and a fall action detection algorithm based on a dual stream convolutional neural network are proposed and a fall action analysis system that can be applied to household scenarios is developed in this thesis.The main work is given as follows:I.A extraction algorithm of human silhouettes based on the optimized GMM is proposed.Firstly,the foreground extraction method based on the optimized GMM is introduced.Then,the initial discrimination problem of the human silhouettes is analyzed,and the contour discrimination method based on the width histogram is proposed.Finally,the extraction of the successive human silhouettes is analyzed,and the contour tracking method based on intersection ratio and center of gravity distance is proposed.II.A fall action detection algorithm based on the dual-stream convolutional neural network is proposed.Firstly,the extraction of the motion history image and the extraction of the motion direction based on MHI is introduced.Then,the dual-stream model architecture is proposed,including the spatial and temporal convolution network structure and fusion judgment.Finally,the TS-Net is proposed,expounding search space,target expectation reward function and search framework.III.A pedestrian fall detection system for home scenarios is developed.Firstly,the system architecture is analyzed,which mainly includes a fall image processing module,a fall detection module,a model training module,and an alarm and distress module.Then,the hardware environment,software environment and implementation process of the system are described.Finally,the effects of the image processing module,the fall detection module,and the alarm and distress module are verified respectively.