Identifying multiple line-structured lights from images via a local-to-global graph representation

Identifying multiple line-structured lights from an image is a fundamental yet challenging issue in the active 3D visual reconstruction. The existing approaches using complex coding schemes are typically time-consuming and inapplicable to real-time sparse 3D reconstruction. In this paper, we solve the multi-line ambiguity from a new viewpoint-distribution pattern of the light segments in the image. We construct a local-to-global graph framework to fully describe the hierarchical distribution of multiple line-structured lights in a 2D image. The lights are firstly grouped as several local graphs according to a light overlapping metric. Then, the hierarchies of the local graphs are unified via the depth of the node, leading to a global graph. The lights in the same level of the global graph come from the same laser plane. The experimental results show the applicability of the proposed algorithm to identify scattered light segments and the robustness to varying sensor poses. We further apply the proposed algorithm to a 3D reconstruction case, achieving a reconstruction precision of 0.025mm. The proposed approach avoids complex auxiliary laser coding and thus is more convenient to conduct.