Scalable Surface Reconstruction from Point Clouds with Extreme Scale and Density Diversity

Abstract

In this paper we present a scalable approach for robustly computing a 3D surface mesh from multi-scale multi-view stereo point clouds that can handle extreme jumps of point density (in our experiments three orders of magnitude). The backbone of our approach is a combination of octree data partitioning, local Delaunay tetrahedralization and graph cut optimization. Graph cut optimization is used twice, once to extract surface hypotheses from local Delaunay tetrahedralizations and once to merge overlapping surface hypotheses even when the local tetrahedralizations do not share the same topology. This formulation allows us to obtain a constant memory consumption per sub-problem while at the same time retaining the density independent interpolation properties of the Delaunay-based optimization. On multiple public datasets, we demonstrate that our approach is highly competitive with the state-of-the-art in terms of accuracy, completeness and outlier resilience. Further, we demonstrate the multi-scale potential of our approach by processing a newly recorded dataset with 2 billion points and a point density variation of more than four orders of magnitude -requiring less than 9GB of RAM per process. * Results incorporated in this paper received funding from the European Unions Horizon 2020 research and innovation programme under grant agreement No 730294 and the EC FP7 project 3D-PITOTI (ICT-2011-600545). 2.5 km 500 m 50 m 50 cm 5 cm arXiv:1705.00949v1 [cs.CV]

Texture super-resolution for 3D reconstruction

“…Malgré l'information essentielle contenue dans la texture, on remarque que de nombreux travaux de reconstruction 3D s'intéressent peu au problème et se limitent à assigner une intensité par primitive du modèle (sommet, facette, voxel, point 3D, surfel). Très souvent, pour représenter un modèle texturé, les travaux utilisent un maillage polygonal et se contentent d'assigner alors une intensité par facette ou sommet [Steinbruecker et al, 2014, Mostegel et al, 2017. Cette façon de procéder à le désavantage majeur de lier la résolution de la texture à celle de la géométrie du modèle reconstruit.…”

Extracting Geometric Structures in Images with Delaunay Point Processes

“…Unfortunately, existing algorithms often decouple the detection of local primitives from the construction of global structures. Continuing on our examples, object contouring methods typically detect line segments along image discontinuities before assembling them to form polygons [1], [2], and multiview stereo reconstruction algorithms extract 3D points by feature matching before interpolating them with a surface mesh [3], [4]. While this two-step approach reduces computational burden, the quality of the resulting structures depends heavily on the local decisions taken during primitive detection.…”

Automatic and Semantically-Aware 3D UAV Flight Planning for Image-Based 3D Reconstruction

“…The mobility and maneuverability of UAVs to freely move in three dimensions and simultaneously capture close-up images of an object with arbitrary viewing angles allow to generate high-resolution and photo-realistic 3D models with high accuracy by processing a series of overlapping images with current state-of-the-art structure from motion (SfM) and multi-view stereo (MVS) pipelines, such as Pix4D [1], Bundler [2], or Colmap [3]. These models are of high interest in various fields, such as the use of digitized building models for 3D city modeling [4], object inspection [5], or cultural heritage documentation [6]. However, the quality of resulting 3D models strongly relies on flight plans that satisfy the requirements of an image-based 3D modeling process which include the acquisition of multiple overlapping images, sufficient baselines between the camera viewpoints and the prevention of optical occlusions from surrounding obstacles.…”