Study On Camera Calibration And Its Correlation Technique

Camera calibration has been one of important topics for photogrammetry, vision inspection, computer vision and so on. It has been useful in many practical applications such as mapping, industry controlling, automatic navigation and military. This paper presents some my researches on subpixel corner detection, camera models and calibration methods that follow in detail.Firstly, a subpixel corner detection method based on Harris corner detector is given. Using iterative method and the corner property that any vector from the true corner to a pixel point in the corner neighborhood is always orthogonal to the gradient vector of the image at the point we obtained corners subpixel coordinates whose precision precedes 0.01 pixel. This solves the problem how to achieve the control points coordinates with subpixel accuracy on camera calibration.Secondly, after the advantanges and the disadvantages of kinds of camera models and camera calibration methods are analyzed, a flexible and high accuracy camera model is established that concerns the lens radial and tangential distortion and the intrinsic parameters uncertainty such as the probable axes skewness of the CCD cells, aspect ratio and principle point decentering.Thirdly, an automatic squares countering method is proposed. The image coordinates of the control points are obtained by using this method and the homography between the planar object and its image. With the aid of the subpixel corner detector mentioned above we solve the matching problem between the control points on the planar pattern in space and their corresponding points on images.Fourthly, an improved Levenberg-Marquardt Algorithm for nonlinear least square problems is proposed which lowers the restriction of the initial values of the parameters for nonlinear iterative method and accelerates the convergence speed. A set of Matlab programs for complete camera calibration basing on moving planar pattern are given, which optimizes the calibration results by using several nonlinear optimization methods and so reduces the pixel error to approximate 0.1 pixel.