利用直线特征进行单幅图像畸变校正方法

A Method for Single Image Distortion Correction Using a Straight Line Feature

针对传统镜头畸变校正方法较少,并考虑畸变中心带来的不利影响,提出了一种基于单幅图像直线特征的畸变校正方法,且能够同步地估计畸变系数和畸变中心。在单幅图像中建立单参数划分模型,深入分析了空间直线与畸变后直线的几何关系,证明了空间三维直线经镜头畸变后,在二维平面形成圆弧,并且从该圆弧圆心为圆心的同心圆上的某点即是畸变中心,进而基于圆拟合方法对畸变参数进行估计。与已有方法相比,该方法考虑了畸变中心的影响,并且直接使用畸变点进行畸变参数估计,因此,具有更高的校正精度和更强的鲁棒性。实验中,首先对拟合图像进行实验,针对不同噪声、不同畸变系数对各种算法展开了三组仿真实验,结果表明,利用列文伯格马夸尔特（Levenberg-Marquardt,LM）算法进行圆拟合和畸变参数估计的方法性能表现良好。展开对真实黑白棋盘的畸变校正实验,运用三种算法对棋盘中13条直线进行畸变校正后的测量,表明LM算法要明显优于其他两种算法,有着良好的畸变校正效果。最后对真实场景的图片进行测试,取得了较为理想的畸变校正结果。

In the traditional lens distortion correction method the adverse effects of distortion center are not taken into account.In this paper,a radial distortion correction method is proposed,in which the distortion coefficient and distortion center can be estimated synchronously.We establish a single image division model and analyze the geometrical relation between linear and straight lines,finding that a 3 Dstraight line forms an arc in the two-dimensional plane after the distortion of the lens,and the center of the circular center is the distortion center.Distortion parameters are estimated based on the circle fitting method.Compared to the traditional method,the proposed method considers the influence of the distortion center,and estimates distortion parameters using the distortion point directly.Therefore,it has higher correction precision and stronger robustness.In an experiment,the fitting image is tested,and three simulation experiments were conducted for different noises and different distortion coefficients on various algorithms.The results show that the method performance of circular fitting and distortion parameter estimation using the LM algorithm is the most effective solution.In a checkerboard calibration experiment,the three algorithms based on the chessboard of 13 lines after distortion correction of the measurement are used.These results show that the LM algorithm performs significantly better than the other two algorithms tested,with a good calibration effect.The image of the real scene was tested and the results of the distortion correction have a relatively ideal effect.