Multi-frame Sample Templet Update Video Images Moving Target Tracking Algorithm Research

Moving target tracking is one of the most important research directions in computervision. Because of the complexity and variability of the environment, moving targettracking does not have a uniform system which can be used in all conditions. Althoughtracking exist some problem, but it has significant research value that drive researcherskeep holding on it. This thesis aimed at tracking that failure caused by moving, pose,occlusion, illumination variation or the other factors like disappear again. We researchedand experimented with different features.First, for fast moving target tracking under the complex background, a dynamicupdate classification of joint multi-frame sample information for motion target trackingalgorithm was proposed. Moving object tracking can be seen as a classification betweentarget and background. Use the first three targets as template, sample foreground andbackground samples, and calculate Haar-like feature values and train the naive Baysclassifier for better performance. The template and feature value will be update during thetracking process. This tracking method can be performed accurate in real time.Secondly, in order to track moving target under large illumination change, weresearched dynamic update integral histogram of multi-frame sample similarity matchingfor moving target tracking algorithm. Traditional histogram accumulates each pixel valueinto corresponding range or bin. This algorithm in this paper considered the entire pixelwith different weights, and then calculated the illumination invariant feature andBhattacharyya Distance for detecting the position of target. It had been proved accurateand effective.Finally, on account of the accident model in target tracking such as occlusion, out ofbounds, and reappear and so on. The tracking method under accident conditions isdiscussed. It chooses the target template in the first frame and shows it with a rectanglebox, then divides it into several blocks for expressing the local information. Calculate thecoefficient of template and each block with the current image use a slide window.Compare the coefficient with a threshold to find the maximum and sign it with a rectangle box. Update the template set online. This algorithm can be performed accurate and robust.