Structure from Motion with Line Segments Under Relaxed Endpoint Constraints

摘要

We present a novel structure from motion pipeline, which estimates motion and wiry 3D structure from imaged line segments across multiple views. Although the position and orientation of line segments can be determined more accurately than point features, the instability of their endpoints and the fact that lines are not constrained by epipolar geometry diverted most research focus away to point-based methods. In our approach, we tackle the problem of instable endpoints by utilizing relaxed constraints on their positions, both during matching and as well in the following bundle adjustment stage. Furthermore, we gain efficiency in estimating trifocal image relations by decoupling rotation and translation. To this end, a novel linear solver for relative translation estimation given rotations from five line correspondences in three views is introduced. Extensive experiments on long image sequences show that our line-based structure from motion pipeline advantageously complements point-based methods, giving more meaningful 3D representation for indoor scenarios.