Research On Algorithms Of Video Target Tracking

Video target tracking technology is a hot research topic in the field of computer vision and image processing.A variety of tracking algorithms were put forward by domestic and foreign scholars in view of the different tracking problems,but which only solve the problem of tracking in a given situation.And,there are many kinds of interference factors,such as the change of target size,the complexity of the scene,the noise and so on.Consequently,a higher requirement for the following algorithm is needed,and it is necessary to study and solve these problems in depth.Aiming at the problems existing in the practical application of video target,the basic research of tracking algorithm is studied in this paper,which is supported by National Natural Science Foundation Project(NO.61201118),and this research draws lessons from the existing classic video target tracking algorithms.The main work and innovation points are as follows:(1)Mean Shift and feature fusion,these two tracking algorithms are introduced,which are two typical algorithms in the current tracking field.In this paper,it expounds the basic theory and implementation process of the two algorithms,and analysis their advantages and disadvantages.(2)A video target tracking algorithm based on the unknown noise variance is proposed.In practical applications,the accuracy of the process noise and observation noise variance can not be obtained.If the noise variance is inaccurate,tracking accuracy is degraded or tracking failure will be brought out.In view of these problems,the extended forgetting factor recursive least squares(EFRLS)method is applied to video target tracking in this paper.First,Mean Shift algorithm is used to obtain the preliminary estimate of the target position.Then the EFRLS method is used to estimate the position in the next frame.Experimental results show that the proposed algorithm is significantly better than traditional Mean Shift and is equivalent to Kalman tracking algorithm combined with Mean Shift.In addition,if severe occlusions exist in between targets,this algorithm is better than Kalman tracking algorithm combined with Mean Shift.The proposed algorithm also has good tracking performance.(3)A video target tracking algorithm based on binary robust invariant scalable keypoints(BRISK)feature matching is proposed.First,the target area of interest is manually calibrated,and BRISK features and descriptors are extracted in the current frame and the next frame.Then,according to a recent neighbor Hamming distance criterion to obtain matching like feature point pairs,and random sample consensus(RANSAC)algorithm is used to remove the wrong matching points.The next step,the coordinates of the center of gravity is calculated separately for the current frame and the next frame,using the correct matching points coordinates of the selected target region.Finally,the target tracking window is established according to the coordinates of the center of gravity,and the rest of the video sequences are treated equally.The experimental results show that tracking real time of the proposed algorithm is better than scale invariant feature transform(SIFT)or speeded up robust features(SURF)under the same conditions.Besides,it has more advantages compared to the traditional Mean Shift algorithm under occlusion circumstances.