Motion Object Detetion And Intelligent Video Surveillance System Design

With the rapid development of computer vision, electronic technology and communication technology, intelligent video surveillance system attracts more and more attention, as it is an efficient way of protecting the security both of body and property. Through analyzing and researching the object detection algorithms, this thesis proposes a responsible, efficient background model and update scheme, and finally uses the scheme to design and implement an intelligent video surveillance system.Motion object detecting is the first step in the process of video processing, and meantime it is difficult and complex in the extent of skills. Motion object detection is a process of analyzing and processing the image sets from video, and the purpose is to monitor the motion objects and analyze their action real-time. Motion object detection can be divided into two aspects: static background and non-static background. This thesis focuses on such a situation: motion detection in static background, and proposes a new background model and update scheme.This thesis consists of three parts: research algorithms, propose a new background model and update scheme and design and implement an intelligent video surveillance system.Firstly, in the extent of implementation mechanism, analyzing and comparing motion detection methods on the foundation of optical flow, background subtraction, frames difference, entropy detection, the thesis lists the advantages and disadvantages of different methods. Next focus on background subtraction in static environment, compare seven background models and update ways: Running Gaussian Average, Temporal Median Filter, Mixture of Gaussians, Kernel Density Estimation (KDE), Sequential KD Approximation, Concurrence of Image Variations, Eigen-backgrounds. Finally aiming to the drawback of traditional ways, such as illumination change, multi-objects in background, image dithering, propose a new motion object detection algorithm to avoid the effect of environment changes. Algorithm testing results show good performance. Using the new background model and update scheme, this thesis designs and implements an intelligent video surveillance system framework. This system can not only detect the exact motion object, but also position face and count the number. As the lowest layer of vision processing, motion object detection is the basis of some advanced applications, and positioning face and counting the number is an embodiment of application.