基于图像特征分布优化的相机外参自标定

External parameter calibration based on image feature distribution optimization

标定两个已知相机之间的相对外参是三维视觉变形测量中的一个基本任务.当使用自标定技术时,一般需要在标定图像中建立立体对应关系,而实际测量场景中复杂的纹理分布对可靠立体特征的建立提出了挑战.为此,本文通过研究图像特征分布的优化策略,提出了一种针对实际测量任务的非约束场景外参标定方法.首先,通过图像特征匹配获得立体图像间的点匹配信息,并利用图像相关算法进行筛选,建立一组可靠的立体点对应关系,从而通过求解本质矩阵得到相机间的外参估计.随后,以重投影误差作为标定精度评估指标,提出一种特征点稀疏和循环补充策略,可对单帧或者多帧图像的特征点分布进行优化,有效实现对标定误差的补偿.实验表明本文的标定算法具有良好的精度和鲁棒性.

External parameter calibration between two known cameras is a fundamental task in three-dimensional visual deformation measurement.The self-calibration technique requires the establishment of stereo correspondences in the calibration images;however,the complex distribution of textures in real scenes poses a great challenge to the development of reliable stereo correspondences.This paper proposes a non-constrained external parameter calibration method for actual measurement tasks by studying the optimization strategy of image feature distribution.First,the correspondences between stereo images are obtained through feature-matching,and an image correlation algorithm is adopted to screen outliers for establishing a set of reliable stereo correspondences.External parameter estimation between two cameras is then achieved by solving the essential matrix.A feature-point sparse and recurrent supplemental strategy using the reprojection error as the metric of calibration accuracy is proposed.This strategy can optimize the feature distribution of single or multiple frames,thereby effectively realizing calibration error compensation.Experiments show that the calibration algorithm proposed in this paper has good accuracy and robustness.