Multi-target 3D Reconstruction from RGB-D Data

The main purpose of this paper is to quickly reconstruct multiple closed three-dimensional models in a scene. In this case, a method for reconstructing objects in a scene interactively is presented. Firstly a 3D scene mesh model with color is reconstructed fromthe RGB-D data. Secondly, a semantic label system is designed to segment different types of surfaces and objects in the scene, which provides a reference for further extraction and repair of models. Finally, an adaptive hole filling algorithm is deployed to recover the original mesh models automatically. It is demonstrated that our method can scan and generate multiple models at once, and the surface quality of the scanned structure is also evaluated.