摘要
为了提高相机标定中特征点提取的精度并减小重建后三维空间的误差,故提出一种基于全阈值分割与多约束的双目鱼眼相机标定算法。该方法针对传统标定过程中边缘区域噪声大、特征点拟合精度不足的问题,首先在已有的鱼眼镜头统一成像模型的基础上,利用高斯函数对灰度直方图进行迭代拟合,通过全局阈值分割方法精确提取标定圆的特征点,实现对畸变图像的高精度特征识别。然后,在反投影求取空间点的阶段,充分利用三维信息,在原有的距离约束与极线约束基础上,进一步引入垂直约束与共线约束,构建了融合多几何约束的优化函数,该函数能够同时考虑整体空间结构与局部几何关系。在实现上,采用圆形标定板结合全局阈值椭圆边界检测策略,并利用Levenberg-Marquardt算法对多约束目标函数进行迭代求解,从而提高参数估计的收敛性与稳定性。实验部分在双目鱼眼系统下完成了多组标定测试,设置了距离误差、极线误差、垂直误差与共线误差等指标进行评估。最终结果表明,所提方法在特征点检测与参数优化方面均具有较高精度,平均重投影误差(MRE)较传统方法降低30.85%,三维空间误差降低约45%,测试图像的平均误差低于0.3%,验证了算法在精度与鲁棒性上的显著提升。该研究为鱼眼双目视觉系统的高精度标定提供了可靠的参量依据。
In order to improve the accuracy of feature point extraction and reduce the error in the reconstructed 3D space in camera calibration,we propose a full-threshold segmentation and multi-constraint binocular fisheye camera calibration algorithm in this paper.To address the problems of large noise in edge regions and insufficient fitting accuracy of feature points in traditional calibration,firstly,based on the unified fisheye imaging model,a Gaussian function is used to iteratively fit the grayscale histogram and a global full-threshold segmentation method is applied to accurately extract circular calibration feature points,thereby achieving high-precision feature identification in distorted images.In addition,multi-constraint optimization function that integrates vertical and collinear constraints is proposed as a supplement to distance and epipolar constraints.It fully exploits 3D information in the inverse-projection stage of spatial point reconstruction,simultaneously considers the overall 3D spatial structure and local geometric relationships of the reconstructed points.In implementation,a circular calibration board combined with a global-threshold-based ellipse boundary detection strategy is adopted,and the multi-constraint objective function is iteratively solved using the Levenberg-Marquardt algorithm,thereby improving the convergence and stability of parameter estimation.Experiments are conducted on a binocular fisheye system,where multiple calibration tests are performed and evaluation metrics including distance error,epipolar error,vertical error,and collinear error were used.The experimental results show that the proposed method attains high accuracy in feature-point detection and parameter optimization:the mean reprojection error(MRE)is reduced by 30.85% compared with traditional methods,the 3D spatial error is reduced by about 45%,and the average error on the test images is below 0.3%,which verifies significant improvements in both accuracy and robustness.This study provides a reliable parameter foundation for high-precision calibration of binocular fisheye vision systems.
作者
刘文昊
宋涛
刘兆伦
产佳
李宁
Liu Wenhao;Song Tao;Liu Zhaolun;Chan Jia;Li Ning(School of Information Science and Engineering,Yanshan University,Qinhuangdao 066000,China;Hebei Key Laboratory of Measurement Technology and Instrumentation,Yanshan University,Qinhuangdao 066000,China;School of Electrical Engineering,Yanshan University,Qinhuangdao 066000,China;Hebei Green&Intelligent Bulk Cargo Large Equipment Intelligent Operation and Maintenance Technology Innovation Center,Yanshan University,Qinhuangdao 066000,China)
出处
《仪器仪表学报》
北大核心
2025年第11期253-259,共7页
Chinese Journal of Scientific Instrument
基金
河北省高等学校科学技术研究项目(CXY2024024)
国家自然科学基金项目(62441113)资助。
关键词
机器视觉
鱼眼相机标定
阈值分割
多约束优化
特征点提取
machine vision
fisheye camera calibration
threshold segmentation
multi-constraint optimization
feature point extraction
