Research On Object Tracking Based On Optical Flow And Contour

Target tracking is a hot research topic in the machine vision,image processing and pattern recognition community.It was widely used in video surveillance,medical diagnosis,robot vision navigation,human computer interaction and intelligent transportation.In complex scenes,it is a challenge task to design an real-time and accurate visual tracking system due to target deformation,camera jitter,light changes,occultation,similar background and ambient noise.In recent years,although many domestic and foreign scholars have carried out deep research in visual about object tracking,and put forward a lot of improved algorithm,but still can't solve the problem of tracking target in real-time and robustly.In this thesis,a deeply research have been done in target tracking based on L-K optical flow with contours and corners.In view of the problem of low real time and can't track appointed object,a target tracking algorithm based on L-K optical flow of adaptive select contour have been presented,which can improve real-time and accuracy.The main innovation research contents and contribution of this thesis are as follows:(1)Aiming at the problem of gray image reflect inaccurately about the trend of three channel's value,an improved method have been present.First,detecting Canny edges in each three channels to gain all three channel images.Then,fusing data of the three images.The test proved that the edges image after fusing data is better than origin edges image.Based on that method,a algorithm of adaptive select contour have been put forward,which to solve the problem of optical flow tracks appointed target inaccurately with the factor of other moving objects,that can reduce the other object reflection and improved accuracy.(2)To solve the issue of the poor real-time of global optical flow,present a method that detect corners that inside contour in further based on the detected target contour and delete other corners,then only tracking corners in contours improve real-time.For the station of little corners result to fail to track,introduce Hu matching to update corners to improved robustness.The experiment result proved the algorithm's availability that the article present.