Research On Object Detection And Tracking Algroithm In Dynamic Background Based On Global Motion Compensation

Target tracking and detection technology of dynamic background is one of theimportant topics in computer vision field. It has a wide range of applications inprecision guided weapon, traffic monitoring, visual navigation of mobile robot andintelligent vehicle, etc. As the background motion caused by camera motion, researchon detection and tracking of moving targets that have important practical significanceand theoretical value.The technology of target tracking and detection of moving multi-target isin-depth studied in dynamic background. This paper is divided into two parts, the firstone analysis of the dynamic of the moving target detecting based on global motioncompensation. And then, target tracking solution based on particle filter is proposed.In the moving target detection part, six-parameters global motion estimation isused based on block-matching to achieve global motion compensation. Eliminatetarget detection and tracking performance against the adverse effects when the camerais in case of non-stability. Image cropping and edge extraction information extractionis taken to image sequences before the block-matching. And by pre-judging macroblock, macro blocks is removed which is not rich in texture information. In theprocess of block-matching search, nine-point-cross search algorithm is proposed.Compared with the traditional search method, the nine-point-cross method can reducethe amount of data and improve real-time of algorithm. Dynamic background'scompensation is made based on the bilinear interpolation after getting global motionparameters of camera. Frame-difference is used in the static background to extract themoving object area. A small target detection method based on morphology processingand the small area denoise is proposed. The algorithm is proved to be effective and itcan extract the dynamic target area in dynamic background.In this paper, weighted histogram color target tracking is adopted based on theparticle filter technology. Though this method, robustness of target tracking isstronger in complex situations, such as target rotation, deformation and non-singlecolor. The weighted histogram color and Bhattacharyya distance are proposed as thetarget color model to describe the similarity between particle and the target colormodel, providing a strong basis to particle weight. The adaptive particle filtering isadopted when select the amount of sample.By further analysis and discussion, it isdemonstrated that the method is effective and practical.