一种鲁棒的基于半正定规划的相机自标定方法

摘要

Camera self-calibration methods use the image data only ,and determine the camera model up to a similarity trans-formation .The absolute dual quadric based linear camera self-calibration method is often used to find an initial solution of the cam-era parameters due to its simplicity and effectiveness .According to the non-positive definiteness of dual image of absolute conic (DIAC ) ,we introduce a semi-definite programming based self-calibration method .It constrains the semi positive definiteness and the validness of intrinsic parameters .The experimental results on both synthetic data and real images verify the effectiveness and ro-bustness of the algorithm .%基于绝对对偶二次曲面的线性自标定方法因为算法简单，并能解决可变参数的自标定问题，常用于为其他非线性自标定算法或捆集调整提供标定初值。针对传统线性自标定方法中反复出现的绝对对偶二次曲线的对偶图像不正定与最小二乘解的局限问题，提出了一种新的基于半正定规划的自标定方法。算法通过在求解过程中约束绝对对偶二次曲面的半正定性与其可行域范围，从而确保能够对DIAC （Dual of the Image of the Absolute Conic ）分解得到内参数矩阵，并使内参数在更合理的范围内取值。通过合成数据和真实图像数据实验，证明了算法的有效性和鲁棒性。