Research On Visual Tracking Algorithm

As a difficult research pot of the computer vision research field, visual tracking has great application prospects such as Video Surveillance, visual reality, augment reality, human-computer-interaction, cartoon motion capture video compression video conference, intelligence transportation and military guidance. At the same time visual tracking is also the basis of some advanced machine vision for example behavior recognitionThis project is a hot point of the international artificial intelligent and the computer vision community a large amount paper is published in all kinds of international conference on machine vision. Different methods are proposed. But because of the complexity of the illumination condition and the object modeling of the real world, no one method is general. In this paper, the project is researched.Tn this paper, the classification of the problem on visual tracking is firstly introduced and then basic procedure to resolve it described. To the static scene condition, this paper focuses on two problems: background modeling and the object shadow removing. To solve the first problem, frame difference information is used to direct the location of object regions of difference image frame, and then the objects location information come from d i f ference frame i s used to compound a background frame that does not include any moving object. To solve the second problem, bright distortion and color distortion are calculated and used to determine whether the region on the object belong to the shadow. At the same time, an adaptive threshold method is proposed. To dynamic scene visual tracking problem, this paper has researched the specific object tracking: face tracking. A structure and color based feature vector has been presented to modeling the face and used to real ize the satisfactory tracking result. To the active tracking problem, a toy car is selected as the research object. The system structure andcontrol method is the main research respect. At the last of the paper, a computer vision based application is introduced. In the application, a linescan based method is used to satisfy the real time and robustness requirement.