采用简化Brown模型及改进BFGS法的相机自标定

发布时间：2018-04-06 21:01

本文选题：相机自标定　切入点：Brown模型　出处：《光学精密工程》2017年09期

【摘要】：为了精确地反映相机的几何成像关系,本文基于简化的Brown模型和改进的BFGS(Broyden-Fletcher-GoldfarbShanno)算法提出了一种相机自标定方法。该方法首先将线性模型和畸变模型拟合为非线性模型,通过线性模型的基本矩阵约束非线性模型参数得到约束方程;然后,提出了适用于非线性内参数约束方程的基于新拟牛顿方程的改进BFGS算法并求解了方程内参数。利用提出的模型和算法,该标定方法能够在较少的迭代次数和有噪声条件下保证标定结果的精度和鲁棒性。有、无噪声情况下的收敛性分析和鲁棒性分析显示:在噪声不大于±3pixel的情况下,迭代10次即能保证重投影误差小于0.4pixel。通过标定相机内参数并计算重投影误差进行了真实图像实验,结果表明:标定精度误差小于0.06%,重投影误差为0.35pixel,验证了提出方法的有效性。该方法适用于计算机视觉领域中的图像处理,模式分类和场景分析等。
[Abstract]:In order to accurately reflect the geometric imaging relationship of the camera, a camera self-calibration method is proposed based on the simplified Brown model and the improved BFGs Broyden-Fletcher-Goldfarb Shannoalgorithm.In this method, the linear model and the distortion model are first fitted into a nonlinear model, and the constraint equations are obtained by the parameters of the basic matrix constrained nonlinear model of the linear model.An improved BFGS algorithm based on new quasi-Newtonian equations for nonlinear internal parameter constraint equations is proposed and its internal parameters are solved.By using the proposed model and algorithm, the calibration method can ensure the accuracy and robustness of the calibration results under the condition of less iteration times and less noise.The convergence analysis and robustness analysis in noise-free case show that if the noise is not greater than 卤3pixel, the reprojection error is less than 0.4 pixel for 10 iterations.The real image experiments are carried out by calibrating the camera parameters and calculating the reprojection error. The results show that the calibration accuracy error is less than 0.06 and the reprojection error is 0.35 pixel, which verifies the validity of the proposed method.This method is suitable for image processing, pattern classification and scene analysis in the field of computer vision.
【作者单位】：中国科学院长春光学精密机械与物理研究所空间机器人中心创新研究室;中国科学院大学;
【基金】：吉林省产业创新专项基金资助项目(No.20160520074JH)
【分类号】：TP391.41

【相似文献】