Research On Moving Object Tracking Technique In Intelligent Surveillance System

Moving object tracking is an attractive research subject of computer vision,which has broad application in many areas. But the loss of information caused by projection of the 3D world on a 2D image, noise in images, complex object motion, illumination changes, partial and full occlusion have all become great obstacles for the research of moving object tracking.Regarding the weakness of traditional tracking algorithm, the paper introduced two new tracking algorithms, which can overcome dynamic background, occlusion and illumination changes.The author extends kernel-based weighted histogram from grey space to HSV space, compensates the original searching location by global motion estimation which raises the possibility of object belonging to the searching window. The new 3-step searching method enhances the searching efficiency in large scale. To overcome fully occlusion, this method processes motion trajectory curve fitting by the method of least squares and estimates the object location through the trajectory obtained while occlusion happens.The object tracking method based on block motion estimation realizes block motion and light compensation by block Harris corner matching, which overcomes sudden light change and get good tracking performance under dynamic background. Based on the thought of multi-feature fuse, combining the color histogram and HOG as the object feature, the algorithm introduces gradient and location information which overcome the problem that the background object and tracking object are similar in color.Experiments show that both of the two object tracking methods achieve real-time and accurate tracking result on video with dynamic background, while the first one can overcome fully occlusion and the second can overcome sudden illumination change.