High-Quality Surface-Based 3D Reconstruction Using 2.5D Maps

Previous works on RGB-D reconstruction are based on voxels, points, or meshes, which are either too computationally expensive to represent high-resolution 3D models, or not convenient for model regularization to deal with noise and errors. In this paper, we propose a new method to reconstruct 3D models with more accurate geometric information and better texture. Our method uses abstract surfaces consisting of different points with similar information as units of the model. To reduce the complexity, we use 2.5D heightmaps to represent each surface in the reconstructed model, making it convenient to do regularization. Experiments demonstrate the effectiveness of our method.