A new method of camera calibration based on the segmentation model

摘要

A new way of camera calibration is proposed in this article, which divides the image into three different regions based on distortion distribution, and different models are used to calibrate each region respectively, then the whole process of camera calibration can be completed. At first, because there is less image distortion in the central region, a linear model is used to complete the calibration in this area. After that, the intrinsic and extrinsic parameters got in the linear calibration are used to calibrate the annular region, in which the influence of the radial and tangential distortion is taken into consideration. Because the points outside the annular area show more unknown distortion, the former distortion mode is not suitable for use any more. Finally, to complete the calibration of this region, the fitting method of smoothing spline is used to get the function of distortion correction. In the experiment, the mean residual of back-projection and normalized calibration error (NCE) are calculated. The results show that the proposed method is quite efficient to improve calibration accuracy compared with traditional methods. It can meet the practical requirement of high precision in vision metrology engineering.