基于虚拟直线的单目视觉坐标测量模型研究

Study on measurement model of monocular vision coordinate based on virtual beelines

在摄像机位置随意固定和测量点共面的约束条件下,基于单目视觉提出了一种基于虚拟直线的单帧图像物点坐标测量的新算法。用几何方法代替常规代数方法,回避了传统的通过求解旋转矩阵和平移向量来获取摄像机位置和姿态参数的过程。利用直线平行、垂直的先验知识,通过空间共线四点经过射影变换的交比不变性得到目标点的位置坐标信息。实验证明:该方法简单、可行,摄像机位置随意摆放,具有很强的可操作性、实用性,且精度较高。

A new algorithm for target coordinates acquisition based on monocular vision is proposed on the condition of camera fixation and co-planarity. The virtual beelines are presented depending on the transformation of projective geometry, which does not require computation camera calibration. Some geometric information is needed to acquire the basic viewpoint invariant cross-ratio, which is an essential invariant with respect to projective transformations. After getting the cross ratio of the 4 points which are collinear, the target coordinate is computed. The experiment result shows that this algorithm is more practical and has high precision as the CCD can be arranged optionally.