Line-based 3D Reconstruction of Wiry Objects

Man-made environments contain many weakly textured surfaces which are typically poorly modeled in sparse point reconstructions. Most notable, wiry structures such as fences, scaffolds, or power pylons are not contained at all. This paper presents a novel approach for generating line-based 3D models from image sequences. Initially, camera positions are obtained using conventional Structurefrom-Motion techniques. In order to avoid explicit matching of 2D line segments in the various views we exploit the epipolar constraints and generate a series of 3D line hypotheses, which are then verified and clustered to obtain the final result. We show that this approach can be used to densify various sparse occupied point clouds of urban scenes in order to obtain a meaningful model of the underlying structure.