Elsevier

Graphical Models

Volume 75, Issue 6, November 2013, Pages 346-361
Graphical Models

Splat-based surface reconstruction from defect-laden point sets

https://doi.org/10.1016/j.gmod.2013.08.001Get rights and content

Highlights

  • The proposed surface reconstruction method is robust to noise and outliers.

  • The method uses raw data point sets to compute a splat-based representation.

  • In a second step, the splats representation is meshed through Delaunay refinement.

  • Neither normal estimation nor the computation of global functions is required.

  • Ransac is used at the two levels to filter outliers.

Abstract

We introduce a method for surface reconstruction from point sets that is able to cope with noise and outliers. First, a splat-based representation is computed from the point set. A robust local 3D RANSAC-based procedure is used to filter the point set for outliers, then a local jet surface – a low-degree surface approximation – is fitted to the inliers. Second, we extract the reconstructed surface in the form of a surface triangle mesh through Delaunay refinement. The Delaunay refinement meshing approach requires computing intersections between line segment queries and the surface to be meshed. In the present case, intersection queries are solved from the set of splats through a 1D RANSAC procedure.

Introduction

The growing variety of scanning devices and technologies nowadays provides measurements of objects in the form of point sets. Despite the advances in scanning technologies, achieving a perfect scan is very unlikely and, in general, these point sets contain two main types of defects: noise and outliers. Noise refers to the inaccuracy of the resulting measurements, and is related to the lack of precision and repeatability of the scanning process used, while outliers are wrong measurements that are produced by errors during the scanning (e.g., reflections on the surface when using laser scanners).

Having the object represented as a point set complicates the interpretation of the data since the notion of visibility of the shape from a given viewpoint is lost. For this reason, surface reconstruction methods are required. We aim at obtaining the surface in the form of a surface triangle mesh, which allows further processing such as remeshing or simplification. In this paper, we present a surface reconstruction method that is able to provide a smooth approximation of the sampled shape while dealing with noise and a large number of outliers. Moreover, the method is able, to some extent, to reconstruct surfaces with boundaries.

Section snippets

Related work

Surface reconstruction methods are broadly divided according to their ability to interpolate or approximate the measured points.

Interpolation-based methods have been proposed mainly by the computational geometry community. They commonly rely on extracting the surface from a space partitioning such as the Delaunay triangulation and its dual, the Voronoi diagram. Well-known methods in this area include the Cocone method [1], and the Power Crust method [2]. They use a principled way to separate

Overview and contributions

Our method is based on computing a global surface approximation using local surfaces. Instead of producing a consistent global representation of the surface through a memory-intensive global solver (such as solving for a signed distance or indicator function) and then using an isosurface meshing approach, our approach performs the merging of the different local surfaces at the meshing step.

We denote as splats these local surfaces, which may not be just planar but higher-degree approximations

Creating the splats

Given the input point set P, we compute a splat-based representation in which each splat is a local approximation of the surface. In its simplest form, a splat is a disk tangent to the surface and with a radius adapted to the local density of the point set. However, our method allows higher-degree approximations through so-called jet surfaces. We next explain how these jet surfaces are computed and how we combine them with RANSAC in order to achieve robustness to outliers.

Meshing

We have generated one splat for each input point where RANSAC was successful, and because of the redundancy between them, the same area may be covered by more than one splat. In order to obtain the final surface triangle mesh from the splats, we use a coarse-to-fine meshing algorithm based on the concept of Restricted Delaunay Triangulation (RDT) and Delaunay refinement [27]. Given a set of points E on or near a surface, the RDT is a subcomplex of the 3D Delaunay triangulation of E formed by

Results

We implemented our method using components from the CGAL library [34]. Our current implementation is sequential. Table 1 lists all the parameters used by the two steps of our algorithm. Note that some parameters are not critical and thus have been set once for all experiments shown. The data-dependent parameters are k (nearest neighbors) and those driving the splat-RANSAC procedure. There are 3 additional parameters required for meshing. All the parameters used to generate the results are

Conclusions and future work

We have presented a surface reconstruction method that is able to recover smooth representations of a surface under a large quantity of outliers and noise. The separation of the splat creation and the surface meshing steps makes the method modular, and provides re-usability of the results obtained at each intermediate step. Our method works without any other information than the raw point set, and the local nature of the individual splats makes the method recover, to some extent, the boundaries

Acknowledgments

The authors thank the Stanford 3D Scanning Repository (Stanford University Computer Graphics Laboratory), Michael Kazhdan (Johns Hopkins University), the AIM@SHAPE consortium (the ISTI-CNR Visual Computing Laboratory and Inria), Renaud Keriven and Jean-Philippe Pons for providing the models used in this paper. This work was partially funded through MINECO under Grant CTM2010-15216, the EU under Grant FP7-ICT-2011-7-288704 and the European Research Council (ERC Starting Grant “Robust Geometry

References (36)

  • B. Li et al.

    Robust normal estimation for point clouds with sharp features

    Computers & Graphics

    (2010)
  • J.-D. Boissonnat et al.

    Provably good sampling and meshing of surfaces

    Graphical Models

    (2005)
  • T.K. Dey et al.

    Provable surface reconstruction from noisy samples

    Computational Geometry: Theory and Applications

    (2006)
  • N. Amenta et al.

    A simple algorithm for homeomorphic surface reconstruction

  • N. Amenta et al.

    The power crust

  • F. Bernardini et al.

    The ball-pivoting algorithm for surface reconstruction

    IEEE Transactions on Visualization and Computer Graphics

    (1999)
  • J. Digne et al.

    Scale space meshing of raw data point sets

    Computer Graphics Forum

    (2011)
  • R. Kolluri et al.

    Spectral surface reconstruction from noisy point clouds

  • P. Labatut et al.

    Robust and efficient surface reconstruction from range data

    Computer Graphics Forum

    (2009)
  • W.E. Lorensen et al.

    Marching cubes: a high resolution 3D surface construction algorithm

    SIGGRAPH Computer Graphics

    (1987)
  • H. Hoppe et al.

    Surface reconstruction from unorganized points

    SIGGRAPH Computer Graphics

    (1992)
  • B. Curless et al.

    A volumetric method for building complex models from range images

  • H.-K. Zhao et al.

    Fast surface reconstruction using the level set method

  • A. Hornung et al.

    Robust reconstruction of watertight 3D models from non-uniformly sampled point clouds without normal information

  • J.C. Carr et al.

    Reconstruction and representation of 3d objects with radial basis functions

  • Y. Ohtake et al.

    3D scattered data approximation with adaptive compactly supported radial basis functions

  • M. Kazhdan et al.

    Poisson surface reconstruction

  • D. Levin

    Mesh-independent surface interpolation

    Geometric Modeling for Scientific Visualization

    (2003)
  • Cited by (10)

    • Out-of-core outlier removal for large-scale indoor point clouds

      2022, Graphical Models
      Citation Excerpt :

      The cluster-based local outlier factor (CBLOF) [22] regards small-clustered points far from the other points as outliers. Campos et al. [23] used a quadratic function to fit the local neighboring points. Then a RANSAC-based procedure is performed to detect the outliers which are far away from the best fitting model.

    • Error-aware construction and rendering of multi-scan panoramas from massive point clouds

      2017, Computer Vision and Image Understanding
      Citation Excerpt :

      Nurunnabi et al. (2014) use the Minimum Covariance Determinant in order to compute a robust estimation of the covariance matrix for the neighbourhood of a point, on which they apply PCA to obtain the normal. Campos et al. (2013) iteratively fit d-dimensional splats (d-jets) to the local neighbourhood of each point in a RANSAC-like loop in order to minimize the impact of outliers. In fact, they do surface reconstruction although normals could be extracted from their estimated splats.

    • Variational Shape Reconstruction via Quadric Error Metrics

      2023, Proceedings - SIGGRAPH 2023 Conference Papers
    View all citing articles on Scopus

    This paper has been recommended for acceptance by Tamal Dey and Peter Lindstrom.

    View full text