High-Precision Robust Calibration of Circular Targets Based on Monocular Vision

Industrial cameras often use the method based on circular target for calibration, because of lens distortion and circular target center feature extraction error, this method has asymmetric projection and noise problems, which easily leads to calibration errors. Aiming at this calibration error, this paper independently designs a uniformly distributed circular target, uses the maximally stable extremal region algorithm (MSER) to extract the features of asymmetric projected posture images with a large amount of data, and obtains the center coordinates with strong robustness after adjustment. In the process of sorting the physical coordinates of the corresponding points of the target, the invariance of cross ratio is used to ensure the robustness of feature matching. The initial calibration result of the center of the adjustment is used as the initial calibration value, and the center coordinates are further iteratively corrected through back projection of the virtual physical target. Experimental results show that compared with the traditional elliptic center extraction method, the re-projection error of the proposed method is reduced from 0.89 pixel to 0.03 pixel, and the re-projection error of the proposed method is reduced from 0.06 pixel to 0.03 pixel compared with the calibration results of the checkerboard target under the same camera position, which realizes the high precision and robust calibration of the circular target.