A robust camera calibration method based on triple orthogonal vanishing points

This paper presents an engineering-oriented robust camera calibration framework for decoupling and estimating camera intrinsic parameters using a structured 3D calibration template. Orthogonal vanishing-point geometry is used as an established initialization constraint, while the main contribution lies in integrating it with multi-line vanishing-point estimation, Mahalanobis-distance optimization, maximum-likelihood refinement, radial-distortion correction, and adaptive error compensation. A Circle Chaotic Golden Sine Algorithm (CGSA) is further introduced to reduce the risk of local optima during parameter refinement. Experimental results show that the inverse-projection error of the calibrated data is within 0.051 mm and that the distance-measurement error obtained using the estimated parameters remains below 0.44 mm, confirming the practical accuracy and robustness of the proposed calibration pipeline under noisy and distorted measurement conditions. An optimizer-controlled ablation on the calibration objective shows that, with the calibration model and error metric unchanged, replacing CGSA with GSA and LM increases the average intrinsic error from 0.93% to 1.07% and 1.29%, respectively.