Research On Target Tracking Algorithm Based On Color Features And SURF Features

Target tracking plays an important role in the fields of human-computer interaction,virtual reality and intelligent monitoring.The key of target recognition and tracking is to select the appropriate features to describe the target.In this paper,we have explored and studied this problem,choosing the feature of color histogram and SURF in the image to describe the target information for tracking the target.In this paper,a method based on the SURF feature detection and matching is integrated into the CamShift tracking algorithm which is based on color feature,for solving the problem of the drift and divergence of the tracking window and the missing of the target.And it can reduce the time cost of the SURF feature matching,too.The algorithm uses the convex packet formed by the SURF feature point to block the target template adaptively,and it assigns different weight coefficients to the target and the background region so that it obtains the weighted H-S two-dimensional color histogram,which improves the shortcoming of the CamShift algorithm which is easily disturbed by the background pixels.The candidate region of the target is obtained by the CamShift algorithm.The SURF feature points are extracted from the candidate region,avoiding the time consuming of extracting feature points on large area background.Then it uses the K-D tree algorithm to match the SURF feature points,and the RANSAC algorithm is used to remove the mismatch points for improving the matching accuracy.The target's affine motion information is calculated by using the matching SURF feature information,and the target is accurately identified.According to the size,position and posture of target,selecting the corresponding template update strategy;estimating the target location and scale as the initial candidate region of the next frame,thus it speeds up the convergence speed of the algorithm iteration.Finally,the tracking results of this algorithm are tested.The experiment shows that the algorithm is more stable and more accurate for target tracking compared with the same kind of improved algorithm,and it can make certain recognition to the target attitude.