The Study On Camera Calibration Method Based On Extend Kalman Filter

Camera calibration is the key stage of recovering 3D shape from 2D image and also is the first step of computer vision. It is widely used in the field of machine vision, virtual reality and other computer image applications. The calibration precision influences computer vision precision directly.The method of camera calibration can be roughly classified into three categories: traditional camera calibration, self-calibration, and camera calibration based on active vision. The precision of traditional camera calibration method is high, but it is based on the quantity of the experiment equipments and the images of target. The precision and robustness of self-calibration are bad. The camera calibration method based on active vision requires knowing the movement of camera exactly, which is very expensive.Aiming at the deficiency of the existed camera calibration methods, a camera calibration method based on Extended Kalman Filter (EKF) is proposed. Taking the image coordinates of the characteristic points and the corresponding world coordinates as the filter's inputs; the estimated value of the camera parameters as the filter's outputs; the optimized values of the camera parameters are obtained through the Extended Kalman Filter.An improved Harris corner detection algorithm base on Hough transforms is given. First using Hough transforms detect the numbers and regions of the corners, then in the specific regions using Harris algorithm detect the corners. This method can solve the imprecise of the corner detection, which is caused by the noise on the images of the target. It is also can solve the threshold question in the Harris corner detection, and extract the corner adaptively. Ultimately we can obtain sub-pixel precision corner coordinates.In order to certify this method, the author designs the simulation and real experiments. The simulation shows that this method is good in robustness. The experiment shows that this method has advantage in calibration precision and operation time when compared with Tsai two-step method and Zhang method.Then the author applies this method in the actual robot vision dynamic online calibration. The result shows that this method is good in flexibility and real-time calibration. The work of this paper makes the preparation for the further research such as 3D reconstruction, visual guidance and so on.