Researching On Algorithms Of Motion Detecting & Tracking In Moving Background

Moving target detecting and tracking technology is a key technology of Computer Vision System. Now, the research on motion detecting and tracking in static background such as video Surveillance system, traffic control system, has already been maturity. The technology has been widely applied to military affairs, traffic, industry, medicine and daily life. Compared with static background, the moving background is the relative motion between moving target and image background because of the motion of camera platform. In moving background, the moving target detecting and tracking has been getting more complex. Especially, the real-time speed and reliability of motion detecting and tracking cannot be fully satisfied in practice. How to exactly distinguish moving target and background is the chiefly task of the technology of motion detecting and tracking in moving background.The problem of motion detecting and tracking in moving background can be analyzed to be described as background motion compensation, moving target detecting and moving target tracking. In this thesis, our study and discuss all focused on the above three sub-problems.The research on background motion compensation is a key problem. After discussing some popular method of feature point detecting, we proposed a background motion compensation method based on SIFT feature. Utilizing the scale-invariant feature, the background motion can be represented effectively. Meanwhile, we do not need to set up any threshold values; we can withdraw the advantage that the feature points of the whole image can be detected validly and fast.We utilize the Hue-Saturation-Value (HSV) color histogram and the Histogram of Oriented Gradients descriptors to encode the visual information of the target's appearance, thus the targets can be tracked more accurately. The researched on moving target tracking is deeply discussed. After studying the Mean Shift algorithm, a target tracking method based on improved Mean Shift algorithm was proposed. Then, we proposed another tracking method which can be used in this framework based on particle filter algorithm, and made the comparison between the two methods.In the end, with the combination of the three techniques, we proposed a system framework which can detect and track target in moving background. Experimental results show that the framework can solve the problem of moving target detecting and tracking in moving background rapidly and effectively.