YUV Color Space And Graph Theory Based The Algorithm Of Shadows-Elimination

As the development of technology and a variety of users'needs, the position of Intelligent Video Surveillance Systems is getting more and more important in the security field. The first step of Intelligent Video Surveillance is to detect and extract the moving object (e.g. people, vehicles,.etc) from the specified monitor scene. Only these regions extract accurately can make the following algorithms, such as Object-Tracking, Object-Identification and Classification.etc, run smoothly. However, in the actual Intelligent Video Surveillance Systems, we can not ignore the influence of the light, moving object always tend to carry some uncertainty shadows, and the moving cast shadows are often misjudged as the object because of its own properties, which affect the subsequent image processing and understanding. Thus, the detecting and elimination of shadows has become a problem worthy to study further.Firstly, this thesis reviews and introduces the development history and research status of Video Surveillance Technology, then analyzes and summarizes the commonly used method for object detection and extraction, focusing on analyses of the frame difference method, optical flow method and the background subtraction, mainly introducing their basic principles, advantages and disadvantages, applicable situation. Finally, we determine to adopt Gaussian mixture model (GMM) method to obtain the moving foreground region. GMM can be proved to have better robustness in the light of changes in the scene, the flag leaf or repeated movement. However, GMM is the segmentation for moving regions and it always extract the regions which contain not only real object but also cast shadow area, so it needs to detect and removal shadows from the result.Considering the accuracy of object extraction is affected by the cast shadows, this thesis analyzes and summarizes the several existing algorithm for shadow elimination. And then, a YUV color space-based and graph-theory-based shadow detection method is proposed to improve the quality of detection. The specific implementation steps as follows:Firstly, the precise seeds of foreground and shadow are obtained by the YUV color space based on the comprehensive consideration luminance(Y) and chrominance information (U, V) and morphological filtering in the foreground region extracted by the GMM. And then, mapping the seed point to a graph, using the graph cut algorithm to obtain the further optimized segmentation of the moving target and shadow. This method weakens the inconvenience of the detection threshold for YUV space, and improving the detection accuracy.Finally, the experimental verification of our method has been done in VC++6.0 environment, with the open source computer vision (OpenCV) library. We choose variety of video surveillance scene to carry out our method, including indoor and outdoor environment, and experimental results show that the proposed method can detect and eliminate the cast shadow effectively.