Research On Some Problem Of Camera Calibration

Camera calibration is widely applicable in image fields, such as computer vision, photogrammetry and 3D reconstruction. Since the camera calibration is an indispensable step to recover 3D information from 2D images and the images have large distortion due to the distortion of cameras' lens, the images often need to undistort. The thesis begins with camera calibration and focus on the two keytechniques about camera calibration--corner detections and camera calibrationalgorithm. The main work can be summarized as follows:Firstly, this thesis investigates the domestic and international researchers' work about camera calibration, especially Z.Y. Zhang's thesis about two steps calibration based on a plane, and realized Zhang's method on Matlab tools.Secondly, to get precise calibration parameters, the first job is to get the image information. Corner is one of the most important information of image. based on Harris's corner detection, a novel method for corner detection using multi-scale corner detection algorithm is researched in the thesis. The wavelet-based formula for measuring the image intensity variation is developed, meanwhile, the auto-correlation matrix is obtained that reflected the scale variation information. The novel method might overcome the drawback that single-scale Harris detector usually leads to either missing significant corners or detecting false corner due to noise.Thirdly, to improve corner detection precision, subpixel corner detection method is researched. According to whether the corner initial values are integer, two new methods are given, which are quadratics approaches and insert value subpixel corner detection algorithm. Through converging the two methods, the experiment proves it might get higher corner detection precision.Fourthly, the thesis improves Zhang's method about two steps calibration based on plane. In the new method, the initial parameters are solved by using the points near the image plane centre. According to the theory of camera distortion, the distortion is very litter near the image plane centre, so the solved initial parameters could be very close to exact ones. At this way, nonlinear optimizes are more easy to converge.Fifthly, to get least reproject errors and solve the nonlinear least square problem, the Newton-Gauss Algorithm is proposed. The new method iterates and optimizes all the parameters. The new method overcomes itsself drawback and fully uses its advantage. In this way, when the tangential distortion is distinct in some lenses, the calibration system could still work.Sixth, the experiment is designed scientifically. All the methods are proved through computer simulation and real experiments. In the final, we analyze and discuss the results and data in the experiment and achieve some helpful conclusions.