3D Reconstruction of Edge Line by ICP-based Matching with Geometric Constraints

This paper presents a novel edge-based 3D reconstruction method using a monocular camera. The edge information is known to be illumination-invariant and to include abundant structural information in a relatively small number of pixels. However, since edge line cannot explicitly determine the pixel-to-pixel correspondence as in the feature point approach, it is difficult to perform accurate matching of pixels in a scene with dense edge lines. In this study, edge-based 3D reconstruction using ICP (iterative closest point algorithm) with geometric constraints has been proposed. In our approach, quasi-rigid body assumption for the edge line deformation and smart search for the matching process are introduced for the improvement of matching robustness in scenes with dense edge lines. Experimental results show that the performance of our method for both motion parallax estimation and depth estimation is greatly improved compared with two recent edge-based methods.