Localization Based On Mobile Robot's Visual System

Vision is important for application of robot, with the improvement of capability of image frame grabber cards, lower price, and the faster development of computer vision, and the most important factor is bright eyes is extremely important for collecting information and intelligence.Robot vision system acknowledges and constructs features, geometrical shape, coordinate and relative position of the objects, and describes, apperceives, explains the real world by visual sensor, and then delivers these information to robot, so, it can help robot's decision system to make decision. Most of robot's vision task is based on vision measurement, which is precondition of treading freely and planning. We must calibrate the intrinsic and extrinsic parameters of binocular measurement system before executing measurement. Currently, measurement of visual system needs measurement precision and dimensional precision of references are very high, isn't property for robot's visual system, especially isn't property for robot's measurement system which has zoomar lens. We need to find calibration and measurement methods that can achieve greater precision and don't need highly precision of the references.My major work is calibrating the binocular measurement system, measuring the distance between binocular robot and objects, i.e. localizing objects and itself, and correcting image distortion.There exists distortion in the source image because of the errors of imaging-model and binocular measurement system. In order to gain more precise results of real-time localization and tracking, we correct second distortion when imaging nearly, and correct radial distortion, tangential distortion and affine distortion when imaging far away, this can minimize distortion of the image and minimize time of computation, improve the precision of calibration.When calculating intrinsic and extrinsic parameters of cameras, because there exists errors using the model of pinhole imaging, and also binocular measurement system, preprocessing image. These errors make calibration's results worse; What's more, the results of equations are odd when control-points are coplanar. We resolve these problems by calibrating with coplanar lines and by importing measurement adjustment algorithm in the context of photogrammetric technology.Then we measure the distance between robot's binocular measurement system and objects by triangular surveying take use of results of calibration, the theory of triangular surveying is that one camera's focus and it's homologous imaging points come into being a radial, another and it's homologous imaging points also come into being a radial, two radials intersect at a point, this point is called object point. We calculate objects' coordinates by the method of least squares. Thus, we achieve real time localization.