Automatic Alignment of 3D Point Clouds to Orthographic Images

Recent progress in structure-from-motion (SfM) has led to robust techniques that can operate in extremely general conditions. However, a limitation of SfM is that the scene can only be recovered up to a similarity transformation. We address the problem of automatically aligning 3D point clouds from SfM reconstructions to orthographic images. We extract feature lines from 3D point clouds, and project the feature lines onto the ground plane to create a 2D feature lines. So we reduce this alignment problem to a 2D line to 2D line alignment(match), and a novel technique for the automatic feature lines matching is presented in this paper.