Camera calibration based on weighted triorthogonal vanishing points

This paper presents a three-step method for decoupling and calibrating camera intrinsic parameters using a 3D calibration template. Combining advantages of traditional and self-calibration methods, it employs Mahalanobis distance optimization and maximum likelihood estimation to suppress feature extraction noise and correct distortion. An adaptive error compensation function improves robustness under noise. The process first derives a closed-form solution from three orthogonal vanishing points, then refines initial parameters and distortion coefficients via nonlinear optimization. Experimental results show that at 0.1-pixel noise level, the relative errors for principal points are only 0.036% and 0.045%. Compared to the DLT algorithm, the method reduces calibration errors for u₀, v₀, f_x, and f_y by 8.8%, 11.9%, 3.5% and 3.5%, respectively.