基于双目视觉的三维测量技术研究

本文对基于双目视觉的三维测量技术研究进行了深入挖掘,主要包括摄像机模型、双目相机的标定、图像校正、立体匹配等。同时借鉴机器学习的分类思想,将角点提取转化为二分类问题,利用随机森林算法来实现角点提取,再利用随机森林的预测结果来实现亚像素级角点提取。该方法相对于传统三维测量中的角点提取算法具有更好的自动化性能,能避免角点集群现象,能实现较高精度的亚像素级角点提取,获得更高精度的二维像素坐标。从而利用高精度的二维像素坐标来获得点的三维世界坐标,这也是基于双目视觉的三维测量技术的基础与核心。

Three-dimensional Measurement Technology Based On Binocular Vision

In this paper, the research of 3D measurement technology based on binocular vision has been deeply explored, including camera model, binocular camera calibration, image correction, feature extraction, feature point matching and so on. At the same time, based on the classification idea of machine learning, the corner extraction is transformed into a binary classification problem, and the random forest algorithm is used to achieve corner extraction, and the sub pixel pixel corner extraction is achieved by random forest predictions. Compared with the traditional 3D corner detection algorithm, this method has better automation performance, and can avoid the phenomenon of corner clusters and achieve high-precision sub-pixel corner extraction to obtain more precise 2D pixel coordinates. The high-precision 2D pixel coordinates are used to get the 3D world coordinates of points. this is also the basis and core of the 3D measurement technology based on binocular vision.