Research Of Abnormal Behavior Recognition Method For The Elderly Based On Video

In today's society,with the aggravation of the aging problem,the pension problem of old solitary people have attracted more and more attention.Therefore,it becomes an urgent needs that we use intelligent technology to establish monitoring system,timely detect abnormal behavior for elderly people.Under this background,this thesis proposed a feature extraction method of the stacked convolution ISA model based on spatial and temporal characteristics in the video data.The extracted features are analyzed by cluster analysis,then the behavior is classified by the support vector machine classifier,and the abnormal behavior is identified.The main research contents of the thesis can be concluded as follow:First of all,the paper summarizes the existing methods of feature extraction,and introduces from the global and local feature extraction.In view of the shortcomings of the existing algorithms,we proposed the stacked convolution ISA model based on spatial and temporal characteristics.The model extracts the temporal and spatial features of human motion based on unsupervised feature learning.It can effectively deal with large sample data and extract more rich and significant features.Secondly,the original learning algorithm has the disadvantages of high dimension calculation and slow convergence speed by the orthogonal constraint,so we propose a new learning algorithm—L-BFGS algorithm,which transforms the orthogonal constraints into unconstrained optimization.The convergence speed is accelerated and the recognition rate is also improved in the improved algorithm.Finally,we introduce the process of abnormal behavior recognition in the elderly.The framework includes the sampling of video samples,data preprocessing,feature extraction,feature clustering and classification.And the MATLAB software was used to simulate the daily video database.