Research Of Visitors Flow Rate Issues In Intelligent Video Surveillance

With the fast development of society and economy, the deployment of public safety video surveillance area has become the normalcy. At present, public areas such as the squares, subway stations, government agencies, large shopping malls have achieved video surveillance, but simply capturing and recording images can not meet the demands. What is more important are video images analysis and processing, aiming to warning and monitoring abnormal situation. Currently, a large number of video monitoring systems need human all-day guard, causing several potential security risks. Hence, how to realize the automatic intelligent analysis of public security video surveillance is the research hotspot in the present studies.In this paper, several technical problems of public security video intelligent analysis based on visitors flow rate have been studied and discussed, the main contents of the thesis include three part below:First, discuss the methods of image foreground extraction. The advantages and disadvantages of several commonly used foreground object detection method have been compared through experimental analysis. Besides, adverse factors such as noises and cavities in images after background-difference foreground detection have been dealt with using morphological method.Second, study the crowd density estimation problem. To begin with, study on analysis methods based on pixels and textural, then extract textural features of crowd taking advantage of Gray-level Co-occurrence Matrix in practical use. In classifying phase, multi-classification methods of decision-making tree SVM based on particle swarm optimization have been proposed to classify the crowd density and achieved good effects, verifying the effectiveness of this method through experiment.Third, analyses the methods of people counting in video images. A new method of people counting which is suitable for a variety of crowd density has been proposed. Taking advantage of fitted curve combined by foreground pixel and number of people to estimate the number of people in low crowd density while using regression method based on ALBP features of images to count the number of people of images in high crowd density, validating the accuracy of the method through experiment.