Feature Based Modeling and Design - PowerPoint PPT Presentation

About This Presentation
Title:

Feature Based Modeling and Design

Description:

Feature Based Modeling and Design Alyn Rockwood Kun Gao KAUST – PowerPoint PPT presentation

Number of Views:80
Avg rating:3.0/5.0
Slides: 61
Provided by: ALYNRO
Category:

less

Transcript and Presenter's Notes

Title: Feature Based Modeling and Design


1
Feature Based Modeling and Design
Alyn Rockwood Kun Gao KAUST
2
Greetings from Saudi Arabia

3
Greetings from Journal of Graphics
ToolsAnnouncing a special issue of Geometric
Algebra and Graphics Applications!

4
Modeling with rectangular patchesProblems
  • current digital tools are unable to decouple
    the creative process from the underlying
    mathematical attributes of the surface. -K.
    Singh
  • Laying out patches
  • Non-intuitive design curves
  • NOT what an artist would choose
  • to represent face.
  • Awkward patch layout

5
What is a Feature?
  • Die Mathematiker sind eine Art Franzosen redet
    man mit ihnen, so übersetzen sie es in ihre
    Sprache, und dann ist es also bald ganz etwas
    anderes.
  • Mathematicians are like Frenchmen if you talk
    to them, they translate it into their own
    language, and then it is immediately completely
    different.
  • Johann Wolfgang von Goethe

6
What is a Feature?
  • Boundaries, G1 discontinuities, creases, high
    curvature regions, ridges, peaks

7
Modeling with features
  • A model that more closely reflects the artists
    conception

8
Desiderata
  • Multisided patches. Feature curves are not always
    rectangular.
  • Freeform topology. Does not constrain design by
    forcing on how to layout patch networks, rather
    than features.
  • Floating curves and points. Interior attributes
    allow fine-tuning and richness with minimal
    input.
  • General curve Input. trig functions and fractals,
    isolated points, derivative information such as
    slopes and curvature enhance modeling effects.
  • G2 continuity. Connecting patches smoothly
  • Functionality. Compact database, rapidly
    computed, analytic surfaces and supports
    rendering.

9
Desiderata
10
A New Approach - foundations
  • Discrete least squares minimizes
  • (x - xi)2 (y - yi)2 (z - zi)2
  • Weighted least squares minimizes
  • wi(x,y,z) (x - xi)2 (y - yi)2 (z - zi)2
  • Where, for example
  • wi(x,y,z) 1 / (x - xi)2 (y - yi)2 (z -
    zi)2 ?i

11
A New Approach
  • In parameter space,
  • find ui, the closest point on the ith footprint
    to given point u

12
A New Approach
  • In parameter space,
  • find di, the distance to closest point on the
    ith footprint

d1
d3
d2
13
A New Approach
  • Footprints are pre-images of features in object
    space via feature maps fi . on the ith footprint.
    Define xi so that fi (ui) xi

14
A New Approach
  • In object space,
  • find weighted least squares solution of the xi
  • where the weights wi(x,y,z) 1 / di

x1
x
x3
x2
15
A New Approach
  • F(u) x

F(u) x
16
A New Approach
  • To summarize
  • From u find point on ith footprint
  • and compute point on attribute
  • to use it in least squares
  • where weight is determined as
  • reciprocal distance.
  • Do it for all footprints

17
Example
x1
x
x3
Move u See x move
u1
u
u3
u2
18
Example
x1
When u is close to the footprint then
the corresponding feature dominates because its
distance is small. The interpolation property
x
x3
u1
u
u3
u2
19
Solving the least squares
  • Let fi (u,v) ? (x,y,z), be the attribute
    functions
  • Let wi (u,v) ? wi ? R be the weight functions.
  • For F (xi(u,v) , yi(u,v) , zi(u,v) )
    minimize
  • E ?i fi(u,v) - F 2 wi(u,v)
  • Hence without loss
  • ?E/?x ?i -2(xi(u,v) - x ) wi(u,v) (xi(u,v)
    - x )2 ?wi(u,v) /?x
  • ?i -2(xi(u,v) - x ) wi(u,v) .
  • Minimizing by setting it to 0
  •   ?i xi(u,v) wi(u,v) x ?i wi(u,v).
  •   implies
  •   x ?i xi(u,v) wi(u,v) / ?i wi(u,v).
  •  
  • Putting it together
  •  

20
Mildly surprising discovery
  • F(u) ? wi(u)fi(u) / ? wi(u)
  • generalizes Shepards formula
  • Weights and interpolants are defined in a
    separate parameter space with general distances

21
Convex combination
  • The weights
  • wi(u) / ? wj(u)
  • sum to 1 (partition of unity).
  • F(u) is affinely invariant.
  • The surface lies within the convex hull of the
    fi(u)
  • The surface reproduces the plane/line.

22
Minimal energy soap film effect
Three lines and a sine curve no connection
needed
23
Higher order continuity
  • Let
  • F(u) S Wi(u)2Li(u)
  • where Wi(u) wi(u)/Sj wj(u) and the loft
  • Li(u) (1-si) fi(ti) si gi(ti).
  • (si and ti are distance and footprint
  • parameter functions of u)
  • F(u) is cotangent to the linear loft Li(u) along
  • the attribute curve fi(ti).
  • Tangency is determined by gi(ti).

24
Interpolation to derivatives
  • Five sided, horizontal slope,
  • varying loft fi(ti) gi(ti)

25
Interpolation to derivatives
  • Five sided and slopes,
  • linear lofts

26
Interpolation to derivatives
  • Five sided and slopes,
  • linear lofts

27
G1 continuity
  • Theorem 1. If F(u) ?i Ri(si, ti) (Wi(u)/ ?i
    Wi(u))2 are defined with
  • separate footprints, where Ri(si, ti) (1- si)
    fi(ti) si gi(ti), then,
  •  ?F(u0, v0)/?u ?Ri(0, t0)/?u and ?F(u0, v0)/?v
    ?Ri(0, t0)/?v  
  • for point (u0, v0) on the footprint at parameter
    t0.
  • Parameter sisi(u,v) is the distance to the ith
    footprint
  • Parameter titi(u,v) is then parametric value of
    the nearest point on the ith footprint
  • Theorem guarantees that if two patches share a
    common curve
  • fi(t) and have two lofts that share tangent
    planes at the common
  • curve, then the surface patches also share common
    tangent planes
  • they are G1 at fi(t).

28
G1 continuity
  • Contouring of three and four-sided
  • patch configuration.
  • Matched lofts
  • across curves and
  • at vertices.

29
Higher order continuity
  • Let
  • F(u) S Wi(u)3Qi(u)
  • where Wi(u) wi(u)/Sj wj(u) and
  • Qi(u) (1-si)2 fi(ti) 2 (1-si) si gi(ti) si
    2 hi(ti) .
  • F(u) is cotangent to the parabolic loft Qi(u)
  • along the attribute curve fi(ti).
  • Curvature is determined by gi(ti) and hi(ti) .

30
G2 continuity
  • Let Qi(s,t) (1-s)2 fi(t) 2(1-s)s gi (t) s2
    hi(t) be a parabolic loft. Consider two such
    lofts for each ith footprint, namely QLi(s,t) and
    QRi(s,t),
  • Theorem 2. Given surfaces L(u) ?i QLi(s,t)
    Wi(u)/ ?i Wi(u) 3, and R(u) ?i QRi(s,t)
    Wi(u)/ ?i Wi(u) 3 in which QLi(s,t) and
    QRi(s,t) meet with G2 continuity on the boundary
    curves of L(u) and R(u), then L(u) and R(u) are
    G2 continuous.
  • Theorem 3. Given surfaces as in Theorem 2 where
    QLi(s,t) and QRi(s,t) meet with twist continuity
    on the boundaries of L(u) and R(u), then L(u) and
    R(u) are twist continuous.

31
G2 Continuity
  • Set of 2, 3, 4 and 5-sided patches
  • Isophote showing curvature continuity

32
G2 Continuity
  • Highly reflective, aesthetic surfaces

33
Editability
  • Multi-sided patches
  • Car with 2-, 3- 4- and 5-sided patches
  • A-pillar a single 7-sided patch

34
Editability
  • Minimal curve input for high expressive content

35
Editability
  • Editting cuves across interior, arbitrary
    parameter position.

36
Editability
  • Editting cuves across interior, arbitrary
    parameter position.

37
Editability auto footprint
  • Footprint space inferred from the shape of the
    patch.
  • Automobile A-pillar
  • 7-sided and lengths
  • and flattening!

38
Editability demo
39
Floating edges
  • Unattached footprint maps to unattached attribute
    curve
  • surface interpolates floating curve

40
Floating edges topology?
  • A topologist is one who doesn't know the
    difference between a doughnut and a coffee cup.
  • John Kelley

41
Floating edges
  • Footprints are orthogonal projections of (red)
    attribute
  • curves. Appalachian mountain trimmed to square
  • (arbitrary, no polygon!)

42
Template parameter spaces
  • Cylindrical footprint space. Circular
    footprints.
  • Distance is is vertical height from point to
    circle.

distance
u
43
Template parameter spaces
  • Similar shapes. Cylindrical footprint space
    reapplied to daffodil- twice.

distance
u
44
Interpolation to fractals
  • Properly defined lofts yields slope of ridge

brown fi(ti), red gi(ti).
45
Two-sided attributes, single patch
  • Floating edge with two lofts.
  • Switch loft on footprint C0 continuity
  • glefti(ti)
  • grighti(ti)

46
Two-sided attributes, single patch
  • Several floating edges (7) with paired lofts.
  • Switch lofts on footprint C0 continuity

47
Two-sided attributes, curves
  • 7 Bezier curves with fractal noise is total data
    base

48
Two-sided attributes, single patch
  • Floating edge with two attribute functions.
  • Switch functions again C-1 continuity!
  • flefti(ti)
  • frighti(ti)

49
Two-sided attributes, single patch
  • Several floating edges (3) with paired lofts.
  • Antelope Island. Cliff

50
Closure of Parametric Curves
  • Attributes include any parametric curve, which
  • allows adding trigonometric noise, for example
  • finew(t) fiold(t) 0, 0.3sin(5pit),
    0.2sin(4pit),

51
Closure of Parametric Curves Trim curves
  • Trimming from footprint (parameter) space to
    object space is traditional (5-sided, noise)

52
Closure of Parametric Curves Watertight merging
of two surfaces
  • Using trim curve as an attribute for 3-sided

53
Closure of Parametric Curves Watertight merging
of two surface
  • Moving 3-sided and trim curve

54
Closure of Parametric Curves Watertight merging
of two surface
  • Emblem on shield
  • and floating curve

55
Operations degree of surface
  • G1 surface with cubic attributes
  • F(u) S Wi(u)2Li(u)
  • For ith term
  • Cubic in t (parameter of attribute)
  • Loft is linear in s (distance parameter)
  • Weight is order (N-1)2 over (N-1)2 in s, where N
    is number of sides.
  • For examples, if N2, 3, 4, and 5
  • then degree in s is 2/1, 5/4, 10/9 and 16/15
  • (s and t are affine maps of u and v.)

56
Operations - order of computation
  • Limit singularity, wi(u) is large number
  • F(u) ?i fi(u) wi(u) / ?i wi(u).
  • Computation is linear with number of attributes
  • Remove singularity
  • F(u) ?i ?j?i 1/ wj (u) fi(ui) / ?i P j?i 1/
    wj (u).
  • Recall wj (u) is reciprocal distance
  • Computation is quadratic with number of
    attributes

57
Economy of data/input
  • Object Patches Curves Lofts Noises
  • Auto body 28 25 25 0
  • Cartoon bird 7 21 21 0
  • Morning glory 2 3 2 2
  • Appalachia 1 6 0 0
  • Pikes Peak 1 7 14 1
  • Daffodil 4 5 4 2
  • A-pillar 1 7 7 0

58
Economy of data/input Palo Duro Canyon captured
with 128 edges
59
Thank you!It is impossible to be a
mathematician without being a poet in soul.Sofia
KovalevskayaPoetry is as exact a science as
geometry FlaubertGeometry is as sublime an art
as poetry - ARSee also www.fredesign3d.com
60
Leifs question
  • Var. Mesh AB surface
  • Piecewise linear Algebraically exact
  • Numerical issues Analytic
  • Mesh database(104-6) Small database(102-3)
  • Vertex/feature editing Attribute editing
  • PDE solutions (flow) None -yet
  • Linear time in vertices Linear in attributes
  • LOD re-initialize LOD add attributes
  • Resolution re-initialize resample
  • None Footprint issues
Write a Comment
User Comments (0)
About PowerShow.com