1 / 12

Normal Estimation for Point Clouds: A Comparison Study for a Voronoi Based Method

Normal Estimation for Point Clouds: A Comparison Study for a Voronoi Based Method. Tamal K. Dey Gang Li Jian Sun (presenting). The normal estimation problem and some existing methods. Problem:

anthea
Download Presentation

Normal Estimation for Point Clouds: A Comparison Study for a Voronoi Based Method

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Normal Estimation for Point Clouds: A Comparison Study for a Voronoi Based Method Tamal K. Dey Gang Li Jian Sun (presenting)

  2. The normal estimation problem and some existing methods • Problem: given a possibly noisy point cloud sampled from a surface, estimate the surface normals from input points • Methods: • Numerical methods: plane fitting [HDD*92] and its variations [PKKG03][MNG04] • Combinatorial methods: Voronoi based [AB99] [DG04, DS05]

  3. Plane fitting method [HDD*92] • Assume the best fitting plane at point p: • Minimize the error term under the constraint • Reduce to an eigenvalue problem:

  4. Weighted plane fitting method (WPF)[PKKG03] • Observation: the best fitting plane should respect the nearby points than the distant points • Define the error term: • Weighting function:

  5. Adaptive plane fitting method (APF)[MNG04] • Consider the points within a ball of radius • Noise assumption mean: , standard deviation: • An optimal radius • Compute in an iterative manner

  6. Voronoi based method • Noise-free Point Cloud [AB99] • The line through p and its pole, the furthest Voronoi vertex of Voronoi cell of p, approximates the normal line at p • Noisy Point Cloud — Big Delaunay ball method (BDB) [DG04, DS05] • The line through p and its pole, the furthest Voronoi vertex of Voronoi cell of p whose dual Delaunay ball is “big”, approximates the normal at p • A Delaunay ball is big if

  7. Normal lemmas

  8. Experimental setup • Add noise to the original noise-free point cloud • The x, y and z components of the noise are independent and uniformly distributed • Noise level • Global scale: the amplitude is a factor (0, 0.005, 0.01, 0.02) of the largest side of the axis parallel bounding box • Local scale: the amplitude is a factor (0, 0.5, 1, 2) of the average distance to the five nearest neighbors • Compute a referential normal from the original noise-free point cloud • Estimation error = • Specially sampled point clouds

  9. Mean error plot

  10. Special Case I: uneven sampling • Sample the surface densely along some curves

  11. Special Case II: the surface with high curvature • A very thin ellipsoid

  12. Summary • In case where the noise level is low, all three methods works almost equally well though WPF gives the best result. • In case where the noise level is high or the sample is skewed along some curves, BDB method gives the best result. • Timing • When #pts ~ million, BDB is safer to use. Otherwise WPF or APF is preferred.

More Related