Camera Calibration Issues Related To Research

Camera calibration is the premise of object spatial information in computer vision, an essential step in three-dimensional reconstruction and an important part of stereo vision. Calibration result affects the accuracy of three-dimensional measurement and the result of three-dimensional reconstruction. At the same time, real-time calibration can meet the needs of navigation in machine vision. Therefore, there are important theoretical significance and practical value in research of camera calibration methods.This paper focuses on camera calibration, summarize analysis and research issues related to the camera calibration. The main contents are as follows:1. As the first step in camera calibration, accurate extraction of corner coordinates will have a huge impact on calibration accuracy. Based on the Harris corner detector, a new corner detection method is proposed. First, this method traverses each pixel of the image to calculate the number of pixels with similar gray value in the local neighborhood, and elect parts of the pixels as corner candidate points. Then the candidate points are extracted in Harris corner detector, which will decrease the range of the Gaussian filtering operation and improve the computing speed. At last, sub-pixel precision corners are extracted to improve the accuracy of corner detection.2. Several camera calibration methods at home and abroad are analyzed, especially, the traditional two-step calibration methods. In the second step of these methods, a nonlinear least squares problem is proposed, whose aim is to find the optimal solution of the calibration parameters by minimize the reprojection error. An improved LM method is suggested. This method can reduce the requirements for the initial value in the early iterations and does not slow down the iteration speed by multiplying a decline factor before former formula. Relative change within the parameters is treated as the basis of convergence, making the algorithm more simple and practical.3. A simulation experiment is designed and carried out in Matlab calibration toolbox, which extract sub-pixel corners in improved Harris algorithm and solve nonlinear least-squares equations in improved LM method. Compared with Zhang method in accuracy and computation time, the experimental results show that the algorithm accuracy is within 0.1 pixels, much higher than Zhang method.4. Self-calibration methods are analyzed. These methods are flexible, but be of low accuracy and robustness. In view of this, a new self- calibration method is suggested. Firstly, the constraints about the intrinsic and extrinsic camera parameters are established by matching the geometric relationship between two or more images. Then the camera parameters can be calculated by genetic algorithm. Finally, two corners of image data are chosen to have an experiment, results show that the calibration results are close to the traditional calibration methods, and the calibration process is much easier than traditional calibration methods.