3D camera pose estimation using line correspondences and 1D homographies

Abstract

This paper describes a new method for matching line segments between two images in order to compute the relative camera pose. This approach improves the camera pose for images lacking stable point features but where straight line segments are available. The line matching algorithm is divided into two stages: At first, scale-invariant feature points along the lines are matched incorporating a one-dimensional homography. Then, corresponding line segments are selected based on the quality of the estimated homography and epipolar constraints. Based on two line segment correspondences the relative orientation between two images can be calculated. © 2010 Springer-Verlag.