| Video object tracking is the critical technology of computer vision and the hotspotof the certain research domain. As a very complicated problem，object tracking problemrelates to many aspects of computer vision research. In recent years, object tracking isused in many applications such as surveillance, perceptual user interfaces, and soon．Among the various tracking algorithms, Mean-Shift tracking algorithm has becomepopular due to its simplicity, efficiency and good performance．As a modeling mechanism based on statistical probability density function,Mean-Shift tracking algorithm is simplicity, efficiency and good performance．First, thetarget can not be tracked when the occurrence of occlusion, the target template is notupdated. Second, when the target’s scale changes, fixed window is limited. Finally, asingle template that models a target area is not conducive to distinguish between thebackground and objectives.This paper focuses on Mean-Shift tracking algorithm. First, Kalman filter iscombined with Mean-Shift algorithm, Kalman algorithm predict the target position, Onthis basis, Mean-Shift algorithm mean shift, then updated Kalman parameters as thetarget location; Second, corner points is extracted by algorithm in two consecutiveframes in the target region, By matching the coordinates of corner points, the affinetransformation relationship is built, affine transformation coefficients is obtained byleast-squares, it is considered as the window width coefficient of variation after kalmanfilter; Finally, we calculate the H component probability histogram of the target area,edge direction histogram of the target area, corner points distribution histogram of thetarget area. Define the regional characteristics of the target and backgrounddifferentiation function. By adaptive integrated of three histograms as a hybridhistogram, the histogram can maximize the distinction between target and backgroundas the feature. Improved algorithm is proved by experiments fully effective. |
PDF Full Text Request