Research On Moving Target Tracking Based On Optical Flow Method

Moving target tracking is using the camera as "eyes" to identify and observe and record moving objects of interest.In recent years it has become a hotspot in research of computer vision.Due to the increasing requirement of the realtime performance and the accuracy of target tracking, related algorithm also need to be constantly improved. In this paper, the optical flow method was used in moving target tracking, concrete study about implementation process of target tracking algorithm has been carried on, and some defects of the optical flow method used in the target tracking was improved.Firstly, studied two classic optical flow method HS method and LK method.LK optical flow calculation method is more suitable for application in target tracking because of its high flexibility and relatively small amount of calculation.For larger interframe displacement, the image pyramid decomposition was used to improve the accuracy of optical flow vectors.Secondly, optical flow vectors solved by LK optical flow method have certain instability. The stability of the solution depends on the window weighting function in the process of computing. It was put forward to introduce the Hessian matrix to improve the LK optical flow computation method in this paper.Before computing optical flow vectors, points within the neighborhood were judged by Hessian matrix. Then a threshold value was setted. External point which is beyond the scope of the threshold value was eliminated.Finally the optical flow vectors were solved with the rest of the reliable points according to the weighted least square method.Then, a moving target was tracked based on optical flow method.The whole process including image preprocessing, corner detection, optical flow computation, optical flow clustering, regional analysis and calibrate the target location.Each frame tracking results was made through simulation experiment.Results showed that the algorithm is sensitive to light, has poor anti-interference ability, and the defects such as the error accumulates over time.Finally, a tracking algorithm with high positioning accuracy based on SIFT feature points matching was introduced.It is put forward to fuse the result of this method and the result of optical flow tracking by Kalman filtering.At the same time of fusion the tracking of optical flow method is reinitialized, and the tracking error is corrected. The experimental results show that the proposed fusion method improves the precision compared with the unimproved algorithm and has the comparative real-time performance with the unimproved algorithm...