Research On Algorithms Of Background Extraction And Moving Object Tracking In Traffic Video Surveillance

The method about background extraction and moving object tracking in video surveillance field, which is widely used in the military, medicine and scientific research fields, is one of the most important researches about computer vision and image coding. The design of background extraction and moving object tracking algorithm directly influence the accuracy and stability of the moving object detection and tracking effect. This dissertation mainly studies two key questions that are background extraction and moving object tracking in video surveillance field. Specifically, this dissertation conducts further researches on the methods of background adaptive modeling, foreground separation and vehicle robust tracking based on fixed camera in video surveillance field.Firstly, this dissertation introduces some digital image preprocessing technologies. These technologies lay solid foundation for background extraction and moving object tracking. The quality of the technologies will directly affect subsequent operations. The digital image preprocessing technologies mentioned in this dissertation mainly include the image contrast enhancement, random noise removal, etc.Secondly, this dissertation conducts detail research on background extraction algorithms, which mainly expound two parts, i.e., the first part introduces the method of background extraction based on codebook, this codebook algorithm adopts a vector quantization and clustering techniques to construct a background model and subtracts the current image from the background model to detect moving object by color and brightness distortion. The second part proposes an improved algorithm of background extraction based on codebook, this new algorithm is effectively improved in two aspects, one is about codeword parameters'setting, the other is about clustering principle. Through the two improved aspects the new algorithm embodies some superiorities, such as fewer parameters in one codeword, fewer threshold value that must decided. Especially clustering principle is greatly simplified in the section that based on gray scale image. Experiments show that this algorithm is more effective than former one mentioned in the first part.Finally, on the research of moving object tracking, some common methods of moving object tracking are analyzed and compared firstly. And then the Mean Shift algorithm is analyzed from the aspects of realization principle and match criterion and search algorithm. At last the Continuously Adaptive Mean Shift Algorithm (Camshift) is mainly discussed in the dissertation. Camshift is a tracking method based on the color character of object. It is a real-time algorithm due to its parameterless feature and fast calculation. Simultaneously, because the color information is the inherent character of the object, using this character to track the moving object is robust. Experiment applied Camshift to the traffic video vehicle surveillance shows that this algorithm can effectively track moving vehicle.