Shape Reconstruction from Samples with Cocone Tamal K. Dey Dept. of CIS Ohio State University
A point cloud and reconstruction
Surface meshing from sample
A point set from satelite imaging
A reconstruction with and without noise
Why Sample Based Modeling? Sampling is easy and convenient with advanced technology Automatization (no manual intervention for meshing) Uniform approach for variety of inputs (laser scanner, probe digitizer, MRI,scientific simulations) Robust algorithms are available
Challenges Nonuniform data Boundaries Undersampling Large data Noise
Nonuniform data
Boundaries
Undersampling
Large data 3.4 million points
Cocone Cocone meets the challenges It guarantees geometrically close surface with same topological type Detects boundaries Detects undersampling Handles large data (Supercocone) Watertight surface (Tight Cocone)
Sampling (ABE98) Each x has a sample within f(x) f(x) is the distance to medial axis
Voronoi/Delaunay
Surface and Voronoi Diagram Restricted Voronoi Restricted Delaunay skinny Voronoi cell poles
Cocone algorithm Cocone Space spanned by vectors making angle /8 with horizontal
Radius, height and neighbors p is the farthest point from p in the cocone. radius r(p): p radius of cocone height h(p): min distance to the poles cocone neighbors N p
Flatness condition Vertex p is flat if 1. Ratio condition: r(p) h(p) 2. Normal condition: v(p),v(q) q with p N q
Boundary detection Boundary (P, , ) Compute the set R of flat vertices; while p R and p N q with q R and r(p) h(p) and v(p),v(q) R:=R p; endwhile return P\R end
Detected Boundary Samples
Undersampling repaired
Holes are created
Tight Cocone Guarantee: A water tight surface no matter how the input is.
Tight Cocone output
Holes are created
Hole filling
Time
Large Data Delaunay takes space and time Exact computation is necessary. Doubles the time. Floating pointExact arithmetic
Large Data (Supercocone) Octree subdivision
Cracks Cracks appear in surface computed from octree boxes
Surface matching
David’s Head 2 mil points, 93 minutes
Lucy million points, 198 mints
Shape of arbitrary dimension
Tangent and Normal Polytopes T (p) = V(p) T(p) N (p) = V(p) N(p)
Experiments
Sample Decimation Original 40K points = 0.4 8K points = K points
Rocker K points Original 35K points
Bunny 0.4 7K points K points Original 35K points
Bunny 0.4 7K points K points Original 35K points
Triangle Aspect Ratio
Medial axis
Noise Outliers Cleaned
Noise (Local) This is a challenge unsolved. Perturbation by very tiny amount is tolerated by Cocone.
Boundaries EngineeringMedical
Geometric Models SportsDrug design
Geometric Models Entertainment Mathematical
Meshing
Boundary Detection
Data set Engine
Undersampling for Nonsmoothness
Modeling by Parts
Simplification Sample decimation vs. model decimation
Guarantees Topology preserved, no self intersection, feature dependent tri3100 tri
Multiresolution tri10202 tri 7102 tri
Model Analysis Feature line detection Detection of dimensionality
Mixed Dimensions
Model Reconstruction after Data Segmentation
Conclusions SBGM with Del/Vor diagrams has great potential Challenges are Boundaries Nonsmoothness Noise Large data Robust simplification Robust feature detection