REMOVAL OF ARTIFICIAL POINTS IN 3-D RECONSTRUCTION

摘要

Fast 3-D reconstruction is often needed in computer vision applications such as robot vision. Most of the stereo approaches are too complicated to be used for this task. To solve this problem, a stereo approach based on the epipolar constraint can be used, but the generation of artificial points is unavoidable. In this paper, an artificial point removal algorithm using space continuity constraint and orientation prediction is presented. First, the artificial point problem in trinocular vision setup is analyzed, and then an artificial point removal algorithm based on the space continuity constraint is de-acribed. To ensure correct initial points, an orientation constraint is proposed. The authors show that the projection of a curve orientation onto an image plane depends not only on the camera pose, but also on the curve orientation in 3-D space itself. According to a tangent invariant theorem, the projection of the curve orientation in the image plane can be estimated from the projection of the curve in the image. With the intersection of two projection planes, the orientation in space can be reconstructed, and thus its projection onto a third image plane can be predicted. Based on the difference between the prediction and the actual curve orientation in the image, correct initial points can be determined. Finally, some experiments demonstrate the validity of the proposed approach.