Projecting points onto a point cloud with noise Speaker: Jun Chen Mar 26, 2008
Data Acquisition
Point clouds 25893
Point clouds 56194
Unorganized, connectivity-free topological
Surface Reconstruction
Noise
Definition of “onto” Close? Which?
Applications Rendering Parameterization Simplification Reconstruction Area computation
References An extension on robust directed projection of points onto point clouds Ming-Cui Du, Yu-Shen Liu (CAD, In press) Parameterization-free Projection for Geometry Reconstruction Yaron Lipman, Daniel Cohen-Or, David Levin, Hillel Tal-Ezer (SIGGRAPH ’07)
An extension on robust directed projection of points onto point clouds Ming-Cui Du, Yu-Shen Liu CAD, In press
About the author ( 刘玉身 ) Postdoctor of Purdue University, Ph.D. in Tsinghua University. 3 CAD, 1 The Visual Computer. CAD, DGP.
Result
Previous work Parameterization of clouds of unorganized points using dynamic base surfaces (CAD, 04) Drawing curves onto a cloud of points for point-based Modeling (CAD, 05) Automatic least-squares projection of points onto point clouds with applications in reverse engineering (CAD, 06)
Weighted squared distances error
Proposition Terminating criterion: Simple, direct
Error analysis (Robustness) True location Independent of the cloud of points
Improved weight distance between p m and the axis stability
Improved weight
Reduce cloud Setting the threshold: 1.
Reduce cloud Setting the threshold: Sort the weights in a decreasing order, then choose the nth weight as threshold. (n=N/100).
References Robust diagnostic regression analysis. Atkinson A, Riani M. (Springer;2000) Robust Moving Least-squares Fitting with Sharp Features Shachar Fleishman, Daniel Cohen-Or, Claudio T. Silva (SIGGRAPH ’05)
Forward vs. backward Backward: Start from the entire sample set, then delete bad samples. Forward: Begins with a small outlier-free subset, then refining by adding one good sample at a time. (robust) Adding of multiple points.
Algorithm 1. Choose a small outlier-free subset Q. 2. The solution is computed to the current subset Q. 3. The point with the lowest residual in the remaining points is added into Q. (Forward) 4. Repeat steps 2 and 3 until the error is larger than a predefined threshold. 5. Compute the projection position for the final Q.
Least median of squares
LMS algorithm
Random sampling algorithm
Robustness P: Probability of success. g: Probability of selecting good sample. k: Number of points are selected at random. (k = p) T: Number of iteration. (T = 1000)
Forward search
Disturbing points
Limitations
Use the first quartile (25%) instead of the median (50%)
Parameterization-free Projection for Geometry Reconstruction Yaron Lipman, Daniel Cohen-Or, David Levin, Hillel Tal-Ezer (SIGGRAPH ’07)
About the author ( Yaron Lipman) Ph.D. student at Tel-Aviv University. His supervisors are Prof. David Levin and Prof. Daniel Cohen-Or. SIGGRAPH, TOG, EG, SGP
About the author (Daniel Cohen-Or) Professor at the School of Computer Science, Tel Aviv University. Outstanding Technical Contributions Award 2005(EG) TOG(19), CGF,TVCG, SGP, VC
About the author (David Levin) Professor of Applied Mathematics, Tel-Aviv University. Major interests: Subdivision Moving Least Squares Numerical Integration CAGD Computer Graphics
About the author (David Levin) Professor of Applied Mathematics, Tel-Aviv University. Major interests: Subdivision Moving Least Squares Numerical Integration CAGD Computer Graphics
Results
Locally Optimal Projection (LOP) θ(r), η(r) are fast decreasing functions.
Regularization
Multivariate L 1 median
Optimization
The iterative LOP algorithm
Theorem If the data set P is sampled from a C 2 -smooth surface S, LOP operator has an O(h 2 ) approximation order to S, provided that Λ is carefully chosen.
Initial guess
Results
Parameters: h
Parameters: μ
Efficient simplification of point- sampled surfaces Mark Pauly, Markus Gross, Leif P. Kobbelt IEEE Visualization, 2002
Particle Simulation 1.Spreading Particles. 2.Repulsion.(SIG.92) 3.Projection.(MLS)
Thank you!