uncertainty and variability in point cloud surface data
DESCRIPTION
Uncertainty and Variability in Point Cloud Surface Data. Mark Pauly 1,2 , Niloy J. Mitra 1 , Leonidas J. Guibas 1. 1 Stanford University. 2 ETH, Zurich. Point Cloud Data (PCD). To model some underlying curve/surface. Sources of Uncertainty. Discrete sampling of a manifold - PowerPoint PPT PresentationTRANSCRIPT
Uncertainty and Variability in Point Cloud Surface Data
Mark Pauly1,2, Niloy J. Mitra1, Leonidas J. Guibas1
1 Stanford University 2 ETH, Zurich
Uncertainty and Variability in PCD
Sources of Uncertainty
Discrete sampling of a manifold Sampling density Features of the underlying curve/surface
Noise Noise characteristics
Uncertainty and Variability in PCD
Uncertainty in PCD
PCD curve/ surface
Reconstruction algorithm
But is this unique?
Uncertainty and Variability in PCD
What are our Goals?
• Try to evaluate properties of the set of (interpolating) curves/surfaces.
• Answers in probabilistic sense.
• Capture the uncertainty introduced by point representation.
Uncertainty and Variability in PCD
Related Work• Surface reconstruction
• reconstruct the connectivity
• get a possible mesh representation
• PCD for geometric modeling
• MLS based algorithms
• Kalaiah and Varshney
• PCA based statistical model
• Tensor voting
Uncertainty and Variability in PCD
Notations
Likelihood that a surface interpolating P passes though a point x in space
)(xPF
Prior for a surface S in MPp(S)
Set of all interpolating surfaces for PCD PPM
Uncertainty and Variability in PCD
Expected Value
Surface prior ?
Characteristic function
Set of all interpolating surfaces ?
PMSxx )p(S)dS(F SP )(
S
S )(S x
xx
0
1
Conceptually we can define likelihood as
Uncertainty and Variability in PCD
How to get FP(x) ?
• input : set of points P
• implicitly assume some priors (geometric)
General idea:
Each point piP gives a local vote of likelihood
1. Local likelihood depends on how well neighborhood of pi agrees with x.
2. Weight of vote depends on distance of pi from x.
Uncertainty and Variability in PCD
Estimates for x
x
x
Interpolating curve more likely to pass through x
Prior : preference to linear interpolation
Uncertainty and Variability in PCD
Likelihood Estimate by pi
p ijiiTij xqp 2))((
High if x agrees with neighbors of pi
Distance weighing
Uncertainty and Variability in PCD
Likelihood Estimates
Normalization constant
N
jijii
Tij pxqpxF
1
2))(()( c
1
ii
Uncertainty and Variability in PCD
Finally…
Covariance matrix (independent of x !)
)()(
)()(
)()(
))(()(
1
1
1
2
xqCxq
xqpppxq
pxqppxq
pxqpxF
iiT
i
i
N
jiji
Tijij
Ti
N
jijii
Tijij
Ti
N
jijii
Tij
i
i
i
ii
c
1
c
1
c
1
c
1
O(N)
O(1)
Uncertainty and Variability in PCD
Likelihood Map: Fi(x)
Estimates by point pi
High likelihood
Pinch point is pi
Uncertainty and Variability in PCD
Confidence Map
How much do we trust the local estimates?
Eigenvalue based approach
• Likelihood estimates based on covariance matrices Ci
• Tangency information implicitly coded in Ci
Uncertainty and Variability in PCD
Confidence Map
denote the eigenvalues of Ci.
3
1
1 /l
liii Low value denotes high confidence
N
iiiiP pxxxC
1
)()(
321iii
(similar to sampling criteria proposed by Alexa et al. )
Uncertainty and Variability in PCD
Confidence Map
confidence
Red indicates regions with bad normal estimates
Uncertainty and Variability in PCD
Noise Model
Each point pi corrupted with additive noise i
• zero mean
• noise distribution gi
• noise covariance matrix i
Noise distributions gi-s are assumed to be independent
Uncertainty and Variability in PCD
NoiseExpected likelihood map simplifies to a convolution.
Modified covariance matrix
convolution
)()(
)(1
)(
1
1
xgxF
dgqCqc
xF
i
N
ii
ii
N
ii
T
ii
P
Uncertainty and Variability in PCD
Scale Space
Better estimates in noisy section
Cannot detect separation
Uncertainty and Variability in PCD
Application 1: Most Likely Surface
Sharp features missed?
Active Contour
Uncertainty and Variability in PCD
Application 2: Re-sampling
Add points in low confidence areas
Given the shape !!
Confidence map
Uncertainty and Variability in PCD
Future Work
Soft classification of medical data Analyze variability in family of shapes Incorporate context information to get better
priors Statistical modeling of surface topology