Research On Image-Based Active Visual Tracking System

Active visual tracking is relative to the tradition visual tracking. The most difference between these two visual tracking is that the camera of the active visual system always keeps moving. The most important characters of the active visual system are self-choice and con-tinue tracking the moving targets from the different directions. The active visual system can be used in many fields, such as UAV target tracking technology, robot control technology, missile tracking etc.In this paper, the considered system is a two-dimension numerical turntable equipped with the industry camera and the laser rangefinder to track a moving target with a fixed rela-tive position and orientation. The two-dimension numerical turntable is controlled to keep the moving target in the center of the image plane. Meanwhile, the real-time distance between the moving target and the camera is retrieved from the laser rangefinder to calculate the target's azimuth and height. In the tracking system, the task of the camera is only to obtain the real-time information of the moving target. However, due to the translation and the rotation of the camera, the background of the image varies in the tracking process, which makes it diffi-cult to detect and track the moving target. How to design a controller to control the translation and the rotation of the camera according to the real-time estimate sates of the moving target, to keep the moving target in the center of the image plane is an another difficult problem.This dissertation makes a further study on the system's model, target recognition, object extraction and the target tracking. The whole paper is organized as follows:First, according to the constructor of the two-dimension numerical turntable, the turnta-ble mathematics model is built based on the Lagrangian method. Image Jacobi Matrixes which describe the relation of the change of the image coordinate and the change of the turn-table's angel and the relation of the change of the image coordinate and the moving target's parameter are built.Secondly, an applicable image process scheme is summarized to detect and estimate the state of the moving targets by analyzing the feature of the image. The image pre-process in-cludes these steps:image's two value arithmetic, median arithmetic etc. So the centroid pixel coordinate can be got by eroding and dilimating the image and using edge arithmetic.Thirdly, a variable universe fuzzy sliding mode approach is applied to keep the moving target in the center of the image plane by combining the variable universe fuzzy control and the sliding model control method. The approach can raise the accuracy of tracking system and ensure its robustness. The simulation and experience both demonstrate the advantage of the variable universe fuzzy sliding mode approach is of high accure, and better than the tradition PID under the same condition.Finally, the real system is built by selecting the reasonable devices in each component according the character of the system and the environment. The real-time control software which includes several Control algorithms is developed in the VC++6.0 enviroment. After testing the real system, the control results of the different controller are shown and compared. Finally, the real-time control demonstrates the advantage of the controller.