COMPUTATIONAL GEOMETRY

INTRODUCTION
PETR FELKEL
FEL CTU PRAGUE
[email protected]
https://cw.felk.cvut.cz/doku.php/courses/a4m39vg/start

Based on [Berg] and [Kolingerova]

Version from 5.10.2016

Computational Geometry
1. What is Computational Geometry (CG)?
2. Why to study CG and how?
3. Typical application domains
4. Typical tasks
5. Complexity of algorithms
6. Programming techniques (paradigms) of CG
7. Robustness Issues
8. CGAL – CG algorithm library intro
9. References and resources
10. Course summary

Felkel: Computational geometry

(2)
1. What is Computational Geometry?
 CG Solves geometric problems that require clever
geometric algorithms
 Ex 1: Where is the nearest phone, metro, pub,…?

Ex 2: How to get there?

[Berg]

Felkel: Computational geometry

(3)
1.1 What is Computational Geometry? (…)
 Ex 3: Map overlay

Copyright: http://webhelp.esri.com/arcgisdesktop

Felkel: Computational geometry

(4)
1.2 What is Computational Geometry? (…)
 Good solutions need both:

– Understanding of the
geometric properties of the problem

– Proper applications of
algorithmic techniques (paradigms) and data structures

Felkel: Computational geometry

(5)
1.3 What is Computational Geometry? (…)
 Computational geometry
= systematic study of algorithms and data structures for
geometric objects (points, lines, line segments, n-gons,…)
with focus on exact algorithms that are asymptotically fast

– “Born” in 1975 (Shamos), boom of papers in 90s

(first papers sooner: 1850 Dirichlet, 1908 Voronoi,…)
– Many problems can be formulated geometrically
(e.g., range queries in databases)

Felkel: Computational geometry

(6)
1.4 What is Computational Geometry? (…)
 Problems:
– Degenerate cases (points on line, with same x,…)
• Ignore them first, include later
– Robustness - correct algorithm but not robust
• Limited numerical precision of real arithmetic ?
• Inconsistent eps tests (a=b, b=c, but a ≠ c)

 Nowadays:
– focus on practical implementations, not just on
asymptotically fastest algorithms
– nearly correct result is better than nonsense or crash

Felkel: Computational geometry

(7)
2. Why to study computational geometry?
 Graphics- and Vision- Engineer should know it
(„DSA in nth-Dimension“)
 Set of ready to use tools
 You will know new approaches to choose from

Felkel: Computational geometry

(8)
2.1 How to teach computational geometry?
 Typical “mathematician” method:
– definition-theorem-proof

 Our “practical” approach:

– practical algorithms and their complexity
– practical programing using a geometric library

 Is it OK for you?

Felkel: Computational geometry

(9)
3. Typical application domains
 Computer graphics
– Collisions of objects
– Mouse localization
– Selection of objects in region
– Visibility in 3D (hidden surface removal)
– Computation of shadows
 Robotics [Farag]

– Motion planning (find path - environment with obstacles)

– Task planning (motion + planning order of subtasks)
– Design of robots and working cells

[Berg]

Felkel: Computational geometry

(10)
3.1 Typical application domains (…)
 GIS
– How to store huge data
and search them quickly
– Interpolation of heights
– Overlap of different data [Berg]
• Extract information about regions or relations between data
(pipes under the construction site, plants x average rainfall,…)
• Detect bridges on crossings of roads and rivers…

 CAD/CAM
– Intersections and unions of objects
– Visualization and tests without need to build a prototype
– Manufacturability

Felkel: Computational geometry

(11)
3.2 Typical application domains (…)
 Other domains
– Molecular modeling
– DB search
– IC design
[Berg]

[Berg]
[Berg]

Felkel: Computational geometry

(12)
4. Typical tasks in CG
 Geometric searching - fast location of :

The nearest neighbor Points in given range

(range query)

Felkel: Computational geometry

(13)
4.1 Typical tasks in CG
 Convex hull
= smallest enclosing convex polygon in E2 or
n-gon in E3 containing all the points

V – set of points

Convex Hull CH(V )

Felkel: Computational geometry

(14)
4.2 Typical tasks in CG
 Voronoi diagrams
– Space (plane) partitioning into regions whose points are
nearest to the given primitive (most usually a point)

Felkel: Computational geometry

(15)
4.3 Typical tasks in CG
 Planar triangulations and space tetrahedronization
of given point set

[Maur]

Felkel: Computational geometry

(16)
4.4 Typical tasks in CG
 Intersection of objects
– Detection of common parts of objects
– Usually linear (line segments, polygons, n-gons,…)

Felkel: Computational geometry

(17)
4.5 Typical tasks in CG
 Motion planning
– Search for the shortest path between two points in the
environment with obstacles

[Berg]

Felkel: Computational geometry

(18)
5. Complexity of algorithms and data struc.
 We need a measure for comparison of algorithms
– Independent on computer HW and prog. language
– Dependent on the problem size n
– Describing the behavior of the algorithm for different data
 Running time, preprocessing time, memory size
– Asymptotical analysis – O(g(n)), W(g(n)), Q(g(n))
– Measurement on real data
 Differentiate:
– complexity of the algorithm (particular sort) and
– complexity of the problem (sorting)
– given by number of edges, vertices, faces,…
– equal to the complexity of the best algorithm

Felkel: Computational geometry

(19)
5.1 Complexity of algorithms
 Worst case behavior
– Running time for the “worst” data
 Expected behavior (average)
– expectation of the running time for problems of particular
size and probability distribution of input data
– Valid only if the probability distribution is the same as
expected during the analysis
– Typically much smaller than the worst case behavior
– Ex.: Quick sort O(n2) worst and O(n logn) expected

Felkel: Computational geometry

(20)
6. Programming techniques (paradigms) of CG

 3 phases of a geometric algorithm development

1. Ignore all degeneracies and design an algorithm

2. Adjust the algorithm to be correct for degenerate cases

– Degenerate input exists
– Integrate special cases in general case
– It is better than lot of case-switches (typical for beginners)
– e.g.:
lexicographic order for points on vertical lines
or Symbolic perturbation schemes

3. Implement alg. 2 (use sw library)

Felkel: Computational geometry

(22)
6.1 Sorting
 A preprocessing step
 Simplifies the following processing steps
 Sort according to:
– coordinates x, y,…, or lexicographically to [y,x],
– angles around point
 O(n logn) time and O(n) space

Felkel: Computational geometry

(23)
6.2 Divide and Conquer (divide et impera)
 Split the problem until it is solvable, merge results
DivideAndConquer(S)
1. If known solution then return it
2. else
3. Split input S to k distinct subsets Si
4. Foreach i call DivideAndConquer(Si)
5. Merge the results and return the solution

 Prerequisite
– The input data set must be separable
– Solutions of subsets are independent
– The result can be obtained by merging of sub-results

Felkel: Computational geometry

(24)
6.3 Sweep algorithm
 Split the space by a hyperplane (2D: sweep line)
– “Left” subspace – solution known
– “Right” subspace – solution unknown
 Stop in event points and update the status
 Data structures:
– Event points – points, where to stop the sweep line
and update the status, sorted
– Status – state of the algorithm in the current position of
the sweep line
 Prerequisite:
– Left subspace does not influence the right subspace

Felkel: Computational geometry

(25)
6.3b Sweep-line algorithm

b
Event types for segments:
- start
a
- end
- intersection

Event points – ordered in event queue

Status: {a}, {a,b}, {c,a,b}, {c,b,a}, …

Felkel: Computational geometry

(26)
6.4 Prune and search

 Eliminate parts of the state space, where the

solution clearly does not exist
< >
– Binary search
prune

– Search trees

– Back-tracking (stop if solution worse than current

optimum)
1
2 5
3 4 6
7 8
9

Felkel: Computational geometry

(27)
6.5 Locus approach
 Subdivide the search space into regions of
constant answer
 Use point location to determine the region
– Nearest neighbor search example

Region of the
constant answer:
All points in this
region are nearest
to the yellow point

Felkel: Computational geometry

(28)
6.6 Dualisation
 Use geometry transform to change the problem
into another that can be solved more easily
 Points ↔ hyper planes
– Preservation of incidence (A œ p fl p*œ A*)
 Ex. 2D: determine if 3 points lie on a common line
A*

C
B p* C*
p
↔
A
B*

Felkel: Computational geometry

(29)
6.7 Combinatorial analysis
= The branch of mathematics which studies the
number of different ways of arranging things
 Ex. How many subdivisions of a point set can be
done by one line?

Felkel: Computational geometry

(30)
6.8 New trends in Computational geometry
 From 2D to 3D and more from mid 80s, from linear
to curved objects
 Focus on line segments, triangles in E3 and hyper
planes in Ed
 Strong influence of combinatorial geometry
 Randomized algorithms
 Space effective algorithms (in place, in situ, data
stream algs.)
 Robust algorithms and handling of singularities
 Practical implementation in libraries (CGAL, …)
 Approximate algorithms
Felkel: Computational geometry
(31)
7. Robustness issues
 Geometry in theory is exact
 Geometry with floating-point arithmetic is not exact
– Limited numerical precision of real arithmetic
– Numbers are rounded to nearest possible representation
– Inconsistent epsilon tests (a=b, b=c, but a∫c)

 Naïve use of floating point arithmetic causes

geometric algorithm to
– Produce slightly or completely wrong output
– Crash after invariant violation
– Infinite loop

[siggraph2008-CGAL-course]
Felkel: Computational geometry
(32)
Geometry in theory is exact
 ccw(s,q,r) & ccw(p,s,r) & ccw(p,q,s) => ccw(p,q,r)

q r

 Correctness proofs of algorithms rely on such

theorems

[siggraph2008-CGAL-course]
Felkel: Computational geometry
(33)
Geometry with float. arithmetic is not exact
 ccw(s,q,r) & !ccw(p,s,r) & ccw(p,q,s) => ccw(p,q,r)

wrong result of the orientation predicate

q r

 Correctness proofs of algorithms rely on such

theorems => such algorithms fail

[siggraph2008-CGAL-course]
Felkel: Computational geometry
(34)
Floating-point arithmetic is not exact
a) Limited numerical precision of real numbers
 Numbers represented as normalized
23 bits stored

±m2e 4 Bytes

52 bits stored

8 Bytes
[http://cs.wikipedia.org/wiki/Soubor:Single_double_extended2.gif]

 The mantissa m is a 24-bit (53-bit) value whose

most significant bit (MSB) is always 1 and is,
therefore, not stored.
 Stored numbers are rounded to 24/53 bits mantissa
– lower bits are lost
Felkel: Computational geometry
(35)
Floating-point arithmetic is not exact
b) Smaller numbers are shifted right during additions
and subtractions to align the digits of the same order
Example for float:
Invisible leading bit – not stored
 12 – p for p ~ 0.5 23 1 Normalized mantisa 23 bit
– 1210 = 11002 = 01000001010000000000000000000000 2

2-1 1
– p = 0.510 = 00111111000000000000000000000000 2
1
– p = 0.500000810 = 00111111000000000000000000001101 2
– Mantissa of p is shifted 4 bits right to align with 12
(to have the same exponent 23)
p = 0.500000810 = 01000001000010000000000000000000 21101
–> four least significant bits (LSB) are lost
– The result is 11.5 instead of 11.4999992
Felkel: Computational geometry
(36)
Floating-point arithmetic is not exact
b) Smaller numbers are shifted right during additions
and subtractions to align the digits of the same order
Example for float:
 12 – p for p ~ 0.5 (such as 0.5+2^(-23) )
– Mantissa of p is shifted 4 bits right to align with 12
–> four least significant bits (LSB) are lost

 24 – p for p ~ 0.5
– Mantissa of p is shifted 5 bits right to align with 24 -> 5 LSB are lost

Try it on [http://www.h-schmidt.net/FloatConverter/IEEE754.html or
http://babbage.cs.qc.cuny.edu/IEEE-754/index.xhtml]

Felkel: Computational geometry

(37)
Orientation predicate - definition
௫ ௬
௫ ௬
௫ ௬

௫ ௫ ௬ ௬ ௬ ௬ ௫ ௫
r
௫ ௬

q
Three points
– lie on common line =0
– form a left turn = +1 (positive)
– form a right turn = –1 (negative)
p
Felkel: Computational geometry
(38)
Experiment with orientation predicate
 orientation(p,q,r) = sign((px-rx)(qy-ry)-(py-ry)(qx-rx))
r = [24, 24]
Ideal return values
dy, + left turn

q = [12, 12]

double
p = [0.5 + dx, 0.5 + dy], dx = k.253
– right turn [0.5, 0.5]
Value of the LSB
p
dx,

Felkel: Computational geometry

(39)
Real results of orientation predicate
 orientation(p,q,r) = sign((px-rx)(qy-ry)-(py-ry)(qx-rx))

Return values during the experiment for exponent > -52

dy, + left turn

Never lie on common line

Predicate returns slightly non-zero

values

– right turn
p

Felkel: Computational geometry

(40)
Real results of orientation predicate
 orientation(p,q,r) = sign((px-rx)(qy-ry)-(py-ry)(qx-rx))

Return values during the experiment for exponent -52

Pivot r Pivot p
Felkel: Computational geometry
(41)
Floating point orientation predicate double exp=-53
Pivot p

[Kettner] with correct coolors

Felkel: Computational geometry

(42)
Errors from shift ~0.5 right in subtraction
 4 bits shift => 24 values rounded to the same value
0 8 16 24 32 40 48 56 64 72 80

 5 bits shift => 25 values rounded to the same value

0 16 32 48 64 80 96

 Combined intervals of size 8, 16, 24,…

0 8 16 24 32 40 48 56 64 72 80 88

Felkel: Computational geometry

(43)
Orientation predicate – pivot selection
௫ ௬
௫ ௬
௫ ௬

4 bits lost 5 bits lost 4 bits lost 5 bits lost

௫ ௫ ௬ ௬ ௬ ௬ ௫ ௫
4 bits lost 4 bits lost

௫ ௫ ௬ ௬ ௬ ௬ ௫ ௫
5 bits lost 5 bits lost

௫ ௫ ௬ ௬ ௬ ௬ ௫ ௫

௫ ௫ ௫

Felkel: Computational geometry

(44)
Little improvement - selection of the pivot
(b) double exp=-53
 Pivot – subtracted from the rows in the matrix

Pivot p Pivot q Pivot r

=> Pivot q (point with middle x or y coord.) is the best

But it is typically not used – pivot search is too
complicated in comparison to the predicate[Kettner]
itself
Felkel: Computational geometry
(45)
Epsilon tweaking – wrong approach
 Use tolerance ε =0.00005 to 0.0001 for float
 Points are declared collinear if float_orient returns
a value ≤ ε 0.5+2^(-23) , the smallest repr. value 0.500 000 06

Idea Reality Reality

Idea: boundary for ε Boundary for ε= 0.00005 Boundary for ε= 0.0001

 Boundary is fractured as before, but brighter

Felkel: Computational geometry [Kettner]
(46)
Consequences in convex hull algorithm

[Kettner04]

p5 erroneously inserted a) p6 sees p4p5 first b) p6 sees p1p2 first

Inserting p6 => => forms p4 p6 p5 => forms p1 p6 p2

Felkel: Computational geometry

(47)
Exact Geometric Computing [Yap]
 Make sure that the control flow in the
implementation corresponds to the control flow
with exact real arithmetic

[siggraph2008-CGAL-course]
Felkel: Computational geometry
(48)
Solution
1. Use predicates, that always return the correct
result -> Schewchuck, YAP, LEDA or CGAL
2. Change the algorithm to cope with floating point
predicates but still return something meaningfull
(hard to define)
3. Perturb the input so that the floating point
implementation gives the correct result on it

Felkel: Computational geometry

(49)
8. CGAL

Computational Geometry
Algorithms Library
Slides from [siggraph2008-CGAL-course]

Felkel: Computational geometry

(50)
CGAL
 Large library of geometric algorithms
– Robust code, huge amount of algorithms
– Users can concentrate on their own domain
 Open source project
– Institutional members
(Inria, MPI, Tel-Aviv U, Utrecht U, Groningen U, ETHZ,
Geometry Factory, FU Berlin, Forth, U Athens)
– 500,000 lines of C++ code
– 10,000 downloads/year (+ Linux distributions)
– 20 active developers
– 12 months release cycle

Felkel: Computational geometry

(51)
CGAL algorithms and data structures

[siggraph2008-CGAL-course]
Felkel: Computational geometry
(52)
Exact geometric computing

Predicates Constructions

orientation in_circle intersection circumcenter

[siggraph2008-CGAL-course]
Felkel: Computational geometry
(53)
CGAL Geometric Kernel (see [Hert] for details)
 Encapsulates
– the representation of geometric objects
– and the geometric operations and predicates on these objects

 CGAL provides kernels for

– Points, Predicates, and Exactness
– Number Types
– Cartesian Representation
– Homogeneous Representation

Felkel: Computational geometry

(54)
Points, predicates, and Exactness

[CGAL at SCG ‘99]

Felkel: Computational geometry
(55)
Number Types
Precission
x
slow-down

[CGAL at SCG ‘99]

Felkel: Computational geometry
(56)
Cartesian with double

[CGAL at SCG ‘99]

Felkel: Computational geometry
(58)
Cartesian with Filtered_exact and leda_real

Number type

… One single-line declaration

changes the
precision of all computations

[CGAL at SCG ‘99]

Felkel: Computational geometry
(59)
9 References – for the lectures
 Mark de Berg, Otfried Cheong, Marc van Kreveld, Mark Overmars:
Computational Geometry: Algorithms and Applications, Springer-Verlag,
3rd rev. ed. 2008. 386 pages, 370 fig. ISBN: 978-3-540-77973-5
http://www.cs.uu.nl/geobook/
 [Mount] Mount, D.: Computational Geometry Lecture Notes for Spring 2007
http://www.cs.umd.edu/class/spring2007/cmsc754/Lects/comp-geom-
lects.pdf
 Franko P. Preperata, Michael Ian Shamos: Computational Geometry. An
Introduction. Berlin, Springer-Verlag,1985
 Joseph O´Rourke: .: Computational Geometry in C, Cambridge University
Press, 1993, ISBN 0-521- 44592-2
http://maven.smith.edu/~orourke/books/compgeom.html
 Ivana Kolingerová: Aplikovaná výpočetní geometrie, Přednášky, MFF UK
2008
 Kettner et al. Classroom Examples of Robustness Problems in Geometric
Computations, CGTA 2006,
http://www.mpi-inf.mpg.de/~kettner/pub/nonrobust_cgta_06.pdf

Felkel: Computational geometry

(61)
9.1 References – CGAL
CGAL
 www.cgal.org
 Kettner, L.: Tutorial I: Programming with CGAL
 Alliez, Fabri, Fogel: Computational Geometry Algorithms Library,
SIGGRAPH 2008
 Susan Hert, Michael Hoffmann, Lutz Kettner, Sylvain Pion, and Michael
Seel. An adaptable and extensible geometry kernel. Computational
Geometry: Theory and Applications, 38:16-36, 2007.
[doi:10.1016/j.comgeo.2006.11.004]

Felkel: Computational geometry

(62)
9.2 Useful geometric tools
 OpenSCAD - The Programmers Solid 3D CAD Modeller,
http://www.openscad.org/

 J.R. Shewchuk - Adaptive Precision Floating-Point Arithmetic and Fast

Robust Predicates, Effective implementation of Orientation and InCircle
predicates http://www.cs.cmu.edu/~quake/robust.html
 OpenMESH - A generic and efficient polygon mesh data structure,
https://www.openmesh.org/

 VCG Library - The Visualization and Computer Graphics Library,

http://vcg.isti.cnr.it/vcglib/

 MeshLab - A processing system for 3D triangular meshes -

https://sourceforge.net/projects/meshlab/?source=navbar

Felkel: Computational geometry

(63)
9.3 Collections of geometry resources
 N. Amenta, Directory of Computational Geometry Software,
http://www.geom.umn.edu/software/cglist/.
 D. Eppstein, Geometry in Action,
http://www.ics.uci.edu/~eppstein/geom.html.
 Jeff Erickson, Computational Geometry Pages,
http://compgeom.cs.uiuc.edu/~jeffe/compgeom/

Felkel: Computational geometry

(64)
10. Computational geom. course summary
 Gives an overview of geometric algorithms
 Explains their complexity and limitations
 Different algorithms for different data
 We focus on
– discrete algorithms and precise numbers and predicates
– principles more than on precise mathematical proofs
– practical experiences with geometric sw

Felkel: Computational geometry

(66)