利用长方体的投影计算摄像机参数

本文以已知长方体作为标准标志物，依据从其一幅投影图象抽取的直线确定7个三角形，以其重心为基础拟合出3条直线，据此确定灭点和角点并使用透视投影基本方程对其中2个角点进行修正，利用这些点构造投影变形矩阵，提出一种计算摄像机参数的简单方法。本文考虑的是三点透视投影。

Estimating Camera Parameters from Projection of a Rectangular Parallelepiped

In robot location, it is necessary to determine the three-dimensional position and orientation of a calibration object relative to the camera from its monocular image.This paper presents a new method for such determination. A rectangular parallelepiped which exists in most scenes is chosen to be our calibration object.The advantages of our method arc as follows: 1. Our calibration object can be in any arbitrary position and orientation relative to the camera without requiring some of the camera parameters to be known beforehand.Even now most authors still require that calibration object be placed in special position and orientation relative to the camera, and also that some of the camera parameters are known beforehand.Zhou et al in 19864] and Chen et al in 19895] waived the requirement that the calibration object be placed in a special way but they still required that some of the camera parameters be known beforehand. In the present paper,it is shown that all camera parameters can be calculated uniquely. 2. Unlike Ganapathy's method 8],our method can estimate the camera parameters uniquely. It is known that constraint in the perspective projection transformation and the orthogonal matrices should be exploited as fully as possible so as to improve estimation. Ganapathy makes partial use of such constraint and his method has gained wide acceptance. But some of Ganapathy's calculation equations are still nonlinear and the solution is not unique. Our method makes full use of such constraint in addition to the fundamental equations of perspective projection. Because full constraint is utilized, all the equations are simple and non-trigonometric and the solution is unique. Because full use of constraint is made,our method also gives accurate estimation even in the presence of disturbance.