交通视频图像的分割方法研究-模式识别与智能系统专业论文.docxVIP

!里型兰丝查盔茎堡圭兰壁垒苎——————垒!!坚竺Abst !里型兰丝查盔茎堡圭兰壁垒苎——————垒!!坚竺 Abst ract Intelligent Transportation System(ITS)is a new means for traffic management，it iS composed of traffic flow detection，vehicle identification，traffic abnormal events detection，and SO on．The real·time detection algorithm of vehicle is the key technique of ITS based On video．As the background of surveillance is complex in the scene， Moving object detection is seriously affected by disturbance and changing illumination．Consequently，it is a challengeable work tO segment the video sequence into semantically meaningful objects． In this thesis，we provided the keystone and basic algorithms of moving object detection with a fixed camera，thereinto，background subtraction is commonly used tO detect changing pixels for the surveillance．The adaptive background updating model is the pivotal technique of background subtraction and the real—time and effective extraction of background affects the detection of moving objects seriously． Accordingly，in this thesis，we focused on acceleration of the extraction algorithm of background、our method can achieve real·time implementation．As color image can provide more visual information，we segmented traffic video and detected shadow in color space，and eventually extracted the moving object accurately and effectively． Our meaningful and detailed research work is organized as follows： (1)Systemically expatiate the keystone of image processing and discuss the basic theory and algorithms of video segmentation respectively，which establishes stability foundation for the posterior study． (2)A fast background extraction algorithm based on Intel SIMD technology and mixture Gaussian model is presented in this thesis．The background extraction algorithm based on mixture Gaussian model can extract the background accurately but slowly，in the modified mixture Gaussian model，the calculation of Jeffrey divergence and the update of the background model have inherent SIMD parallelism and can be optim