| With fast improvement of the transport infrastructure in modern city, thetransport load is growing rapidly, which results in a significant increase in the rate oftraffic accidents. Considering the high pressure of current transport system, establishingan intelligent video surveillance system is imperative. Intelligent transport system is afrontier subject of electronic information when applied in the transport video surveillancesystem. It is difficult to locate objects relying solely on two-dimensional video that isemployed in most current transport surveillance system. The introduction of stereo vision,especially binocular vision in transport video surveillance system can significantly solvethe above mentioned problems which are difficult to handle in two-dimensional videosurveillance system. Depth information from stereo vision can help accurately locate theposition of targets. Transport video surveillance system shall be improved to a high levelby applying stereo vision.For3D video surveillance, analysis of moving object is critical, including thedetection, localization and tracking of moving target, as well as the image enhancementin video sequence. The detection in plus with extraction of moving object is importantlyfundamental for the other processing technologies with the object. Shadow handling ofthe moving object to get accurate target area is crucial. After that, we can extractappropriate feature of the object in the stereo images for stereo matching and subsequenttarget localization and tracking. Meanwhile, since the target area is accurately identified,quality, temporal and spatial scalability video coding algorithm can be employed toenhance the target region as well as highly focused area in the video sequence. Theultimate purpose is to achieve better quality of the content in the region of target andhighly focused area in the video surveillance sequence with general temporal and spatialresolution, in condition that the internet bandwidth is constrained. Current algorithms in related to target detection, localization, human tracking, andvideo image enhancement are summarized in the thesis. Based on this, the feasibility andeffectiveness of these algorithms in3D video transport surveillance system are furtherstudied. The main works in the thesis are as follows:1. Moving object detection and extraction algorithm in video surveillance system isstudied. A combination of methods with Gaussian Mixture Modeling of color image,shadow detection and removal in HSV color space as well as contour tracking with levelset method is proposed to detect and extract the moving target.2. Camera geometry model and the principle for stereo matching are analyzed. Theposition of the target in3D video sequence is located based on stereo matchingtechnology.3. The algorithm of Annealed Particle Filter (APF) is introduced after a survey ofcurrent methods for human body tracking. In this chapter, we proposed an articulatedhuman motion tracking algorithm with improved silhouette and view adaptive fusion forhuman tracking in3D video surveillance system.4. The technique of scalability video coding scalable enhancement technology isstudied. We propose a transmitting method based on region of interest has been proposedfor quality scalability video coding by using MGS technology in SVC Firstly, videoimages are artificially segmented into two partitions, ROI and background. Subsequently,for ROI, the16DCT coefficients of the residual, which are acquired by subtracting baselayer from enhance layer, are transmitted hierarchically to the enhance layer. Byemploying the quality scalability coding technology of MGS, the quality of ROI isgradually enhanced. For the background, the16DCT coefficients are controlled not totransmit to the enhance layer, resulting that only base layer is assigned to the region. Byemploying this technique, better quality of the content in ROI can be achieved incondition that the internet bandwidth is constrained. |
PDF Full Text Request