A Comparative Study of Some 3D-2D Computer Vision Algorithms

摘要

3D to 2D analysis and calibration of transformation parameters and structural data in camera centred coordinate frame are fundamental problems in Computer Vision. In this paper we present two novel linear algorithms to calibrate 3D-2D structural and transformation parameters. The algorithms use 9 unknowns to represent the rotation matrix, assuming that correspondence information about a 3D scene and a 2D image is available. The two algorithms are compared with a well known 2D-2D calibration algorithm proposed by Tsai and Huang which is extended to the 3D-2D case. The comparative analysis shows that the proposed algorithms are generally robust and accurate and that they can be used either with further rigid constraints or fine-tuned into robust matrix computation algorithms.