Study On Human Abnormal Activity Recognition Based On Video Image

The study of human abnormal activity recognition has important practical significance in terms of security engineering construction,video surveillance images of real-time analysis is to match the behavior database by computer,if the human behavior is judged to be abnormal behavior,active intervention and warning will happen automatically to protect the safety of life and property.This thesis research includes:firstly,Gaussian mixture model is improved as a background model to model the human movement background and optimize the parameters of Gaussian mixture model,When background updates,a threshold value is detemined to process the case of being misjudged of the prospect.Bayesian decision criteria is improved foreground segmentation to process the case of being misjudged of the prospect.In the shadows and noise elimination aspect,RGB is translated into HSV so that the shadow is eliminated by improved HSV;Secondly,the tracking method integrated the advantage of Kalman filter algorithm and Mean shift algorithm is proposed to track the human motion;Finally,the features of moving target in the image data is extracted,Hidden Markov Model is used for training and classification to identify human abnormal behavior.Experimental results of the proposed method show that the proposed algorithm can identify smashed cars?wandering and other human abnormal behavior recognition effectively.