A Study Of Human Action Recognition Technology Based On Artificial Neural Network

In recent years, fighting acts of violence attracted more and more widespread concern of the whole society with the gradual increase in social instability. Such as banks, prisons, elevators, railway stations, airports and other places have installed a variety of monitoring equipment. Nevertheless, in the process of using traditional video surveillance equipment, for the case of many unexpected violence, fights, fighting, robbery and etc, we can only through the video to find clues, because of the monitoring personnel can’t always guarantee every move occurred among the surveillance video. At this time the violence has occurred, people’s lives and property have been subjected to unlawful infringement; more role of surveillance video is to make up afterwards. Such an action would highlight many problems of traditional video surveillance; a lot of violence will be avoided if we can achieve automatic intelligent detection and alarm in the first time of the violence. This article aims the phenomenon of different types of human behavior in public places, makes a research of human behavior recognition technology among the intelligent video surveillance. Extraction of moving targets can be divided into three researches, including Extraction and segmentation of moving object, image feature extraction and classification of image target. This article researches and analyzes existing algorithms about intelligent video, proposed an improved background subtraction approach to make target segmentation, using principal component analysis for image feature extraction, using artificial neural networks for recognition of human behavior in videos and realized it Initially. Specific studies include:1. Designed improved background differencing, Introduced the background change factor, we can constantly update the background in the process of doing background differencing, because of the role of coefficients. Thereby we can reduce the unrelated background as much as possible at the time of target segmentation.2. By using ten representative body motion videos in Weizmann video library such as running, jumping, waving and etc as a research material. After segmenting the Different human video images for background and foreground, extracting feature vectors for subsequent analysis by principal component analysis technique, reduces the dimension of the feature vector significantly, thus avoid the danger caused by the dimensions of disaster.3. In the comprehension of behavioral characteristics, this article proposes to recognize human features by using artificial neural network. This paper reduce the difference between the output value and the true value continuously by establishing multi-layer artificial neural network and using steepest descent method, which seeks its gradient by sigmoid function constantly. Achieve the purpose of self-learning ultimately.