Uncalibrated Reconstruction Algorithm Based on Angle Restriction and Single Image for an Active Machine Vision System

In research on machine vision, the theory and technique of recovering 3D scenes are important research subjects. They have been developed from 3D reconstruction method with calibrated cameras to uncalibrated cameras. Generally speaking, in order to reconstruct a 3D scene, at least three images must be needed. These images are grabbed at three different viewpoints with a moving camera keeping its internal parameters constant. In this paper, however, it is carried out in metric space with an uncalibrated camera and a single image of the scene. It is based on active machine vision system including a color encoded pseudo-random array pattern projector. In the paper, we focus on the uncalibrated 3D reconstruction algorithm for a scene. The algorithm is divided into three steps: projection reconstruction, affine reconstruction and Euclidean reconstruction up to a scale.