基于双目立体视觉的岩体结构三维建模及其参数优化

3D construction and optimal parameters of rock mass structures based on stereo vision

摘要:

为研究岩体结构面的快速、高效建模方法,基于便携型LenaCV双目立体视觉成像系统,对左、右双目图像进行校正、对齐、纠偏和三维建模算法研究,提出确定相机内外参数的最优标定工况,分析适于点云建模的目标对象特征,确定三维点云建模的最佳拍摄距离与基线距离关系,获取双目视觉建模的最优参数。研究结果表明:以0.5像素为误差上限,LenaCV相机标定的最佳棋盘格尺寸为10 mm、最优标定距离为25倍基线距离；对比三维点云模型建模效果,表面粗糙对象建模效果优于光滑对象,目标在视野中图幅占比高的对象建模效果优于图幅占比低的对象；12 cm基线距离相较于6 cm基线距离的LenaCV双目系统,建立并输出的目标对象的三维点云数据结果更优。基于几种现场测试讨论了最优标定工况、摄影工况和建模对象特点,为岩体三维结构重建提供了可参考的实践经验,验证了基于LenaCV双目摄像机的三维岩体结构面建模的可行性。

Abstract:

To study the fast and efficient modelling method of rock structures, calibration and 3D point cloud establishment were conducted to investigate the optimal parameters of binocular vision modelling through images correction, alignment, deskew and 3D modelling algorithms based on the LenaCV binocular system. Optimal parameters for binocular vision modeling were determined according to the characteristics of target objects suitable for point cloud modeling. Relationship between the optimal shooting distance and baseline distance for 3D point cloud modeling was discussed. The results show that the optimal grid size of the checkerboard is 10 mm. The optimal calibration distance is 25 times the baseline distance with an error limit of 0.5 pixels. Objects with rough surfaces have better 3D modelling effect than that with smooth surfaces. Objects with higher proportion of targets in the field of view show better 3D point cloud modeling than lower ones. The LenaCV binocular system with a baseline distance of 12 cm gives better results than that with a baseline distance of 6 cm in creating and outputting 3D point cloud. Based on on-site tests, the optimal calibration conditions, photographic conditions, and object characteristics for modeling were discussed, providing practical experience for the reconstruction of three-dimensional structures of rock masses and verifying the feasibility.