Estimation of the rigid transformation between two cameras from the Fundamental Matrix VS from Homographies.

摘要

The 3D reconstruction is an important step for the analytical calculation of the Jacobian of the image in a process of visual control of robots. In a two-camera stereo system that reconstruction depends on the knowledge of the rigid transformation between the two cameras and is represented by the rotation and translation between them. These two parameters are the result of a calibration of the stereo pair, but can also be retrieved from the epipolar geometry of the system, or from a homography obtained by features belonging to a flat object in the scene. In this paper, we make an assessment of the latter two alternatives, taking as reference an Euclidean reconstruction eliminating image distortion. We analyze three cases: the distortion inherent in the camera is corrected, without corrected distortion, and when Gaussian noise is added to the detection of features.