CS3500

Computer Graphics
Module: 3D Graphics
P. J. Narayanan
Spring 2005

CS3500
 1

3D Graphics: Objectives
• Representation and manipulation of 3D objects built using
3D primitives.

• Placing a camera in the 3D world and computing what it

sees.

• Final output is a 2D image. A 3D-to-2D projection is

involved.

• Additional complication: Which object is in the front and

which is at the back?
CS3500 December 09, 2004
 2

2D Graphics
• Draw on a 2D canvas using 2D primitives.
• View it frontally from a certain distance away.
• Only scaling and translation between canvas coordinates
and frame buffer coordinates.
• A window on the canvas is mapped to a window on
screen or frame buffer using the (window-to-)viewport
transformation.
• Translate to origin, scale to new size, translate to new
position.
CS3500 December 09, 2004
 3

Perspective Effects
• Perspective distortion is relevant for 3D points.

• Farther points are scaled by larger numbers.

• Distances from the camera are important as everything

happens with respect to the camera.

• First, compute the world with respect to the camera.

• Second, project the 3D world to a 2D plane.

CS3500 December 09, 2004

 4

3D Coordinates
• Cartesian: (x, y, z).
Z
• Polar: (ρ, θ, φ)
z
• z = ρ cos θ,
y = ρ sin θ sin φ P

x = ρ sin θ cos φ θ
ρ
y
Y
φ
• ρ2 = x2 + y 2 + z 2 ,
x
φ = tan−1(y/x),
p
−1
θ = tan ( x2 + y 2/z) X

CS3500 December 09, 2004

 5

Homogeneous Representation
• Convert to a 4-vector with a scale factor as fourth.
(x, y, z) ≡ [kx ky kz k]T for any k 6= 0.

• Translation, rotation, scaling, shearing, etc. become

linear operations represented by 4×4 matrices.

• What does [x y z 0]T mean?

• [a b c d]T could be a point or a plane. Lines are specified

using two such vectors, either as join of two points or
intersection of two planes!
CS3500 December 09, 2004
 6

Transformation
 Matrices: Rotations
1 0 0 0
0 cos θ −sin θ 0
 
Rx(θ) =  ,
 
0 sin θ cos θ 0
0 0 0 1
   
cos θ 0 sin θ 0 cos θ −sin θ 0 0
0 1 0 0 sin θ cos θ 0 0
   
Ry =   , Rz = 
 
−sin θ 0 cos θ 0  0 0 1 0

0 0 0 1 0 0 0 1

• CCW +ve; orthonormal; length preserving; rows give

direction vectors that rotate onto each axis; columns ....
CS3500 December 09, 2004
 7

Translation, Scaling, Composite

   
1 0 0 a a 0 0 0
0 1 0 b 0 b 0 0
   
T (a, b, c) =  , S(a, b, c) = 
0 0 1 c 0 0 c 0

0 0 0 1 0 0 0 1

• A sequence of transforms can be represented using a

composite matrix: M = RxTRy ST · · ·

• Operations are not commutative, but are associative.

• RT and TR??

CS3500 December 09, 2004

 8

Rotation Followed by Translation

   
a a
 
 I b R 0 R b
   
=
 
c  0 1 c
  
  
0 1 0 1

   
a t · r1
 
R 0  I b   R t · r2
  
=

0 1  c t · r3
 
  
0 1 0 1

CS3500 December 09, 2004

 9

Commutativity
• Translations are commutative: T1T2 = T2T1.

• Scaling is commutative: S1S2 = S2S1.

• Rotations are NOT commutative: R1R2 6= R2R1.

• Consider the effect on Z-axis of Rx(90)Ry (90) and

Ry(90)Rx(90)!

• Also, R1T2 6= T2R1 etc....

CS3500 December 09, 2004

 10

Rotation about an axis α

• What is Rα(θ)?

• 3-step process:

1. Apply Rαx to align α with the X axis.

2. Rotate about X by angle θ.
1
3. Undo the first rotation using R−
αx .

1
• Net result: Rα(θ) = R−
αx Rx(θ)Rαx

CS3500 December 09, 2004

 11

β
α Y

γ
X P

CS3500 December 09, 2004

 12

Computing Rα
• First rotate by −β about X axis. Vector α would lie in the
XY plane, with tip at point P.

• Next rotate by −γ about Z axis. Vector α would coincide

with the X axis.
√ 2 2 √
αz αy +αz 1−α2x
• Also, tan β = αy and tan γ = αx = αx if |α| = 1.

1
• Overall: Rαx = Rz(−γ)Rx(−β) and R−
αx = Rx (β)Rz (γ).

• Could have aligned α with Y or Z axis with same results.

CS3500 December 09, 2004
 13

Example: Rotation about [1 1 1]T

(0, 1, 1)

(1, 0, 1) (1, 1, 1)

(1, 1, 0) (1, 2 , 0)

CS3500 December 09, 2004

 14

Computing Rαx: Method 1

 
1 0 0
• Rotate by −π/4 about X. RX(− π4 ) =  0
 √1 √1 
2 2 
0 −1
√ √1
2 2
 1 
√ √2 0
√ 3
 −2 √1 6
• RZ(− arctan( 2)) =  √ 0

6 3

0 0 1  1 
√ √1 √1 0
3 3 3
 
√ −2
√ √1 √1 0 
• Rαx = RZ(−tan−1( 2)) RX(−π4 ) =  6 6 6
 

 0 −1
√ √1 0 
 2 2 
0 0 0 1

CS3500 December 09, 2004

 15

Alternate Way to get Rαx

• After rotation, α will align with X-axis. Hence that is the
first row r1.

• Find an axis orthogonal to α to be the second row. Take

an arbitrary vector v (not parallel to α). Now, r2 = α × v.
 
α 0
v × α 0
 
• Lastly, r3 = r1 × r2 and Rαx = 
 
 α×v×α 0 

0 1

• Many possibilities, all with the same result!

CS3500 December 09, 2004
 16

Computing Rαx: Method 2

• [1 1 1]T will lie on X-axis. First row r1 = [ √13 √1 √1 ]T.
3 3

• Second row r2 = α × [1 0 0]T = [0 √1 √−1 T

] .
2 2
 
√1 √1 √1 0
3 3 3
 
 0 √1 −1
√ 0  √
• Thus, Rαx =  −2 2 2
 = RY (arctan( 2))RX( π4 )
 
 √ √1 √1 0 
 6 6 6 
0 0 0 1

CS3500 December 09, 2004

 17

Rαx Method 2: Contd

(1, 0, 2) (0, 1, 1)

(1, 0, 1) (1, 1, 1)

(1, 1, 0)

CS3500 December 09, 2004

 18

Rotation: Arbitrary Axis, Point

• An arbitrary axis may not pass through the origin.

• Translate by T so that it passes through the origin.

• Apply Rα.

• Translate back using T−1.

• Composite transformation: T−1 Rα T.

CS3500 December 09, 2004

 19

End of Class 7

CS3500 December 09, 2004

 20

3D Transformations
• Many ways to think about a given transform.

• Ultimately, there is only one transform given the starting

and ending configurations.

• Different methods let us analyze the problem from

different perspectives.

CS3500 December 13, 2004

 21

Another Example
Z Z

B
A
C
B’
Y A’ Y
C’

X X

CS3500 December 13, 2004

 22

Working the Example

• Translate by -A to bring it to the origin.
• After the rotation, AC sits on the X axis.
• The first row of the rotation matrix is AC / |AC|.
• The vector normal to the plane ABC sits on the Y axis.
• The second row of the rotation matrix is the unit vector
along AB × AC = (AB × AC) / |AB × AC|.
• Third row is a cross product of the first two.

CS3500 December 13, 2004

 23

Transforming Lines
• A composite transformation can be seen as changing
points in the coordinate system.
• These transformations preserve collinearity. Thus, points
on a line remain on a (transformed) line.
• Take two points on the line, transform them, and compute
the line between new points.
• Lines are defined as a join of 2 points or intersection of 2
planes in 3D. The same holds for transformed lines using
transformed points or planes!

CS3500 December 13, 2004

 24

Transforming Planes
• A plane is defined by a 4-vector n (called the normal
vector) in homogeneous coordinates.
• The plane consists of points p such that nTp = 0.
• Let Q transform n when points are transformed by M.
• Coplanarity is preserved: (Qn)TMp = 0 = nTQTMp.
• True when QTM = I, or Q = M−T.
• Q is the Matrix of cofactors of M in the general case when
M−1 doesn’t exist.

CS3500 December 13, 2004

 25

A Model
Building
WC
Floor2
M M
M Room1 Room1
M
Box
M M
Box M M M

Chair Table Chair1 Chair2 Table

CS3500 December 13, 2004

 26

End of Class 8

CS3500 December 13, 2004

 27

3D Graphics: Block Diagram

• Objects are specified in their own coordinate system and
placed in the world coordinate frame.
• Camera is also placed in the world coordinate frame.
• Camera-to-world geometry is first projected to normalized
coordinates and then to screen.

View View Clipping

Modelling
Orientation Mapping Perspective
Object World Camera Normalized Screen
Coords Coords Coords Coords Coords

CS3500 January 03, 2005

 28

3D Graphics: Block Diagram

Obj 1
Modlng 1
WC VRC
Obj 2
Modlng 2 Viewing

Obj N
Modlng N Projection

NC
Window Coords Viewport

CS3500 January 03, 2005

 29

Different Coordinates
• Object Reference: Object is described in an internal
coordinate frame called ORC.
• World: Common reference frame to describe different
objects, called WC.
• Camera/View Reference: Describe with respect to the
current camera position/orientation, called VRC.
• Normalized Projection: A standard space from which
projection is easy, called NPC.
• Screen: Coordinates in the output device space.
CS3500 January 03, 2005
 30

Transformations
• Modelling: Convert from object coordinates to world
coordinates (ORC to WC).
• View Orientation or Viewing: From world coordinates to
camera coordinates (WC to VRC).
• Simple coordinate transformations.
• View Mapping or Projection: From VRC to Normalized
Coordinates (NC).
• Viewport: From NC to window coordinates.

CS3500 January 03, 2005

 31

Modelling and Viewing

• Transform points from object coordinates (ORC) to world
coordinates (WC) to camera coordinates (VRC).

• A series of transformations for each object or point.

• PVRC = V M PORC.

• Points move to new coordinates in the same global

coordinate system when going from right to left.

• Coordinate system undergoes transformation with points

getting new interpretations when going from left to right.
CS3500 January 03, 2005
 32

A Single Table
• A table defined in its own coordinate system.

CS3500 January 03, 2005

 33

Many Tables in a Room

• Place many tables in a world coordinate system.

CS3500 January 03, 2005

 34

Modelling
• Goal: Transform object coordinates to world coordinates.

• Method: Place ORC fame in the world coordinate frame.

• A single transformation matrix or modelling matrix with

translation, rotation, scaling.

• A unit cube at origin can generate any cuboid using

translation/rotation/scaling.

• Different objects have different modelling matrices.

CS3500 January 03, 2005

 35

Example: Aircraft in a Polar World

Z
• WC frame on ground, yaw
ORC frame on the Y

aircraft. roll
pitch
• Controllers think in X
Z’
polar coordinates for
ρ θ
position and roll-pitch-
yaw for orientation. Y’

• What are the modelling

steps? φ X’

CS3500 January 03, 2005

 36

Aircraft in a Polar World

• Start with both axes aligned.
• Translate to the location given by (ρ, θ, φ).

• Apply yaw, pitch, and roll in that order.

• Coordinate axes undergoing transformation!

• Net effect: T(ρ, θ, φ) Rz(y) Rx(p) Ry (r).

• T(ρ, θ, φ) = Rz(−φ)Ry (θ)T(0, 0, ρ)Ry (−θ)Rz(φ).

CS3500 January 03, 2005

 37

Hierarchy of Transformations
• A hierarchy of transformations needed to setup the world
and the camera.
• A humanoid robot could have a coordinate frame on its
body, another one on the shoulder, a third on the shoulder
that moves with the upper arm, a fourth on the elbow, a
fifth on the elbow that moves with the forearm, etc.
• Remember the wheel with an ant moving on its spoke!

• M = T1 T2 T3 · · · captures the composite transform as

a shift in coordinate frames.
CS3500 January 03, 2005
 38

Humanoid Robot

CS3500 January 03, 2005

 39

Modelling Different Objects

World Coord
Frame

Modelling Modelling Modelling

Matrix 1 Matrix 2 Matrix N

Object Object Object

1 2 N

CS3500 January 03, 2005

 40

Scene Graph

T T

T T T

T T T T

• Objects organized hierarchically with transforms.

CS3500 January 03, 2005

 41

Building
WC
Floor2
M M
M Room1 Room1
M
Box
M M
Box M M M

Chair Table Chair1 Chair2 Table

CS3500 January 03, 2005

 42

Modelling in OpenGL
• OpenGL has a single ModelView matrix M that
transforms points given by glVertex.

• Right multiplication or Mp is applied.

• glTranslate, glRotate also right-multiply M.

• glTranslate command is same as: M = M T.

Similarly for others.

• Modelling commands are: glTranslate, glRotate,

glScale, ...
CS3500 January 03, 2005
 43

• M = T1 R2 S3 R4 translates to the program fragment:

glTranslate()
glRotate()
glScale()
glRotate()

CS3500 January 03, 2005

 44

End of Class 9

CS3500 January 03, 2005

 45

View Orientation or Viewing

• Placing the camera in the world and orienting it right.

• Has 6 degrees of freedom: 3 for position and 3 for

orientation.

• Goal: Transform points expressed in WC to VRC.

• Viewing Transformation can be specified in many ways.

• Commonly using: Camera location, Look point, and Up

direction.
CS3500 January 06, 2005
 46

Change of Coordinate Frame

2
P
2
• PG = (2, 2) PB = (5, 4). 2
3
• T(3, 2) aligns Blue to
Green.
• PB = T(3, 2) PG.
2

CS3500 January 06, 2005

 47

Viewing Specification
• Camera center specified in world coordinates.

• Look point or vector and Up vector known in WC.

v

n
Z Up
C
Look
u

WC Y
X

CS3500 January 06, 2005

 48

Transformation Steps
How do we align WC to VRC?

• Translate to C = (x, y, z).

• Rotate to align Z-axis to -(Look Vector).
• Rotate to align Y-axis to Up.
• Translation is easy. How do we get the rotation matrix?
• Remember columns of the matrix give directions to which
the axes rotate!!

CS3500 January 06, 2005

 49

Rotation
• Let l̄ = L̄/|L̄| and t̄ = Ū/|Ū| be the unit vectors in those
directions.

• Third column of the matrix: n̄ = −l̄.

• Up vector needn’t be orthogonal to the look vector. The L̄

and Ū vectors define the Up plane.

• First column: ū = t̄ × n̄/|t̄ × n̄|

• Second column: v̄ = n̄ × ū.

CS3500 January 06, 2005
 50

View Orientation Transformation

    
x 0 x
 I y   ū v̄ n̄ 0   ū v̄ n̄ y
    
0
V = =

z 0 z
 
    
0T 1 0 0 0 1 0 0 0 1

• We have achieved: PWC = V PVRC.

• But, ..... we needed the reverse!

0
• Viewing transform: V = V −1 = RT T(−C).

CS3500 January 06, 2005

 51

Viewing from an Aircraft in Polar

World
Z
yaw
Y
• Need to give the roll
pilot’s view from pitch
X
aircraft. Z’

ρ θ
• What are the viewing
Y’
steps?

φ X’

CS3500 January 06, 2005

 52

Aircraft in Polar World: Viewing

• Start with both axes aligned.
• Translate to the location given by (ρ, θ, φ).
• Apply yaw, pitch, and roll in that order.
• Coordinate axes undergoing transformation!
• Net effect: T(ρ, θ, φ) Rz(y) Rx(p) Ry (r).
• But, we want camera coordinates from world coordinates!
• Viewing transform: Ry(−r) Rx(−p) Rz(−y) T−1(ρ, θ, φ).
• Inverse of modelling or placing aircraft in WC.
CS3500 January 06, 2005
 53

Modelling and Viewing in OpenGL

• Modelling and Viewing are not truly independent.

• What ultimately matters is only the relative geometry

between the camera and the object(s).

• OpenGL uses only a single ModelView matrix to

represent the relative geomtery.

• Other graphics toolkits may use separate matrices for

conceptual clarity:

CS3500 January 06, 2005

 54

Setting up Objects and Camera

• First arrange all objects in WC. (Modelling).
• Next place the camera in WC. (Viewing)
• Thus, viewing commands are issued first, followed by
modelling commands!
• In OpenGL, commands for both are same:
glRotate(), glTranslate(), glScale().
• Camera at origin, looking along −Z with Y axis as the up
vector on start when ModelView matrix is identity.

CS3500 January 06, 2005

 55

Identity as ModelView Matrix

Y

v
u X
n

Z
CS3500 January 06, 2005
 56

Structure of an OpenGL Program

// Camera is given by Pos & Orientation in WC.
glLoadIdentity(); // Default starting point
glTranslate(−Pos); // Move WC away from VRC
glRotate(−Orient); // Orient WC away!
// WC is set. Model each object with it as reference.
glPushMatrix(); // Save for later use
glRotate(r1); // Start transformation1
glTranslate(t1); // and so on
... Object Geometry ... // Actual polygons/objects
glPopMatrix() // End of object1. Go to next

CS3500 January 06, 2005

 57

Polar View

• A camera always looks at an object. Its position is

specified using polar coordinates.

• Distance d, azimuth az measured from Z-axis ([0, 2π]),

twist tw about camera’s own axis ([0, 2π]), and elevation
el measured from the horizontal (X-Z) plane ([− π2 , + π2 ]).

• How do we align WC to VRC? (We want the inverse!)

CS3500 January 06, 2005
 58

el

az

CS3500 January 06, 2005

 59

Polar View: OpenGL code

(i) Rotate by −az about Y

(ii) Rotate by −el about X
(iii) Rotate by tw about Z
(iv) Translate by d along Z
glTranslate(0, 0, −d);
glRotate(−tw, 0, 0, 1);
el
glRotate(el, 1, 0, 0); az
glRotate(az, 0, 1, 0);

CS3500 January 06, 2005

 60

Modelling & Viewing: Summary

• Place objects in the world coordinate frame

• Place camera in the world coordinate frame

• Can compute object points in camera coordinate frame

• PVRC = V · M · PORC

CS3500 January 06, 2005

 61

End of Class 10

CS3500 January 06, 2005

 62

Discussion & Examples on Modelling/Viewing by the TA

CS3500 January 10, 2005

 63

End of Class 11

CS3500 January 10, 2005

 64

January 13th: Holiday (Sankrathi/Bhogi)

January 17th: Mid-semester examinations

CS3500 January 10, 2005

 65

Projections
• Projection involves projectors starting from 3D points and
hitting the 2D projection plane, forming the image of the
point.

• Two types of projections.

• Parallel projection: Projectors are parallel to each other,

all have the same direction of projection.

• Perspective projection: All projects pass through a point

in space called the centre of projection (COP).
CS3500 January 20, 2005
 66

Parallel Projections
• Orthographic: Projection plane is perpendicular to the
direction of projection.
? If direction of projection parallel to the axes:
plan, elevation, side elevation.
? If DoP (±1, ±1, ±1): isometric projection.

• Oblique: Otherwise.
? Cabinet when projectors make 45 degrees with the
projection plane.
? Cavalier when they make arctan(2) degrees with the
projection plane.
CS3500 January 20, 2005
 67

Oblique Projections
• Projector angle φ and angle
with horizontal α.
• Cabinet: Length along the
depth axis preserved.
φ
• Cavalier: Length along depth α
axis halved. More realistic.
• α is between 30 and 45
degrees.
• Orthographic when φ = 90.

CS3500 January 20, 2005

 68

Orthographic Projections

• Lengths parallel to the projection plane are preserved.

• Only direction of projection matters; distance from the

point to the projection plane doesn’t.

• Good approximation for a camera with a long focal length.

• Plan, elevation, side views etc.

CS3500 January 20, 2005
 69

CS3500 January 20, 2005

 70

Orthographic Projection Equation

• Can be expressed as a matrix equation:
    
x 1 0 0 0 X
y 0 1 0 0  Y 
    
 =
  
 z  0 0 0 0  Z 
 
w 0 0 0 1 1

• If uniform scaling is involved, the top two 1’s should be

the scale factor.

CS3500 January 20, 2005

 71

Perspective Projections
• Can be characterized by
the number of vanishing
points. (projections of points
at infinity).
• Depends on the number of
axes the projection plane
intersects.
• 1-point, 2-point, and 3-point
perspective projections.

CS3500 January 20, 2005

 72

Geometry of Perspective Projection

xp yp
• d = xz , d = yz , z = d.
• In matrix form,
    
x 1 0 0 0 X
 y  0 1 0 0  Y 
    
 = x
 z  0 0 1 0  Z 
 
w 0 0 d1 0 1 xp

• Coordinates scaled down d z

proportional to the depth or
z values.

CS3500 January 20, 2005

 73

Another View
• If we shift the origin to lie on the projection plane, with the
CoP at (0, 0, −d), we get the matrix:
    
x 1 0 0 0 X
 y  0 1 0 0  Y 
    
 =
 z  0 0 0 0  Z 
 
w 0 0 d1 1 1

• We can now think of the Orthographic Projection matrix

as the special case when d → ∞.

CS3500 January 20, 2005

 74

Projections: Summary
   Perspective  
x 1 0 0 0 X   Orthographic
  
 y  0 1 0 0  Y 
     x 1 0 0 0 X
 =  y  0 1 0 0  Y 
    
 z  0 0 1 0  Z 
 
 =
 z  0 0 0 0  Z 
 
w 0 0 d1 0 1
w 0 0 d1 1 1
    
x 1 0 0 0 X
A special case of
 y  0 1 0 0  Y 
    
 = perspective projection
 z  0 0 0 0  Z 
 
with CoP at infinity.
w 0 0 d1 1 1

CS3500 January 20, 2005

 75

Volume of Visibility
• Cameras have finite fields of view in horizontal and
vertical directions.

• What is the shape of its visible space?

• A cylinder for orthographic projections and a cone starting

from the CoP for perspective may seem natural.

• Mathematics is difficult for cones; rectangular structures

are easier!

• View Volume: The volume of potentially visible space.

CS3500 January 20, 2005
 76

View Volume

• View volume is a cube for orthographic cameras and a

(truncated) pyramid for perspective projections.

• 4 planes (left, right, top, bottom) define the view volume.

• Graphics cameras use 2 additional planes to limit

visibility: near & far!

• Planes are specfied in VRC; they move with the camera.

CS3500 January 20, 2005
 77

f
b
l
n
r
Orthographic View Volume
CS3500 January 20, 2005
 78

f
b
l

r
n
Perspective View Volume

CS3500 January 20, 2005

 79

What About the Focal Length?

• An ideal pin-hole camera has the whole world in focus.

• Finite focal-length lenses introduce the effect of focus in

real cameras.
• Even for them, the depth of field (region in focus)
increases as the f-stop increases or the aperture gets
smaller.
• Computer Graphics simulates ideal pin-hole cameras.

• Depth of field can be simulated by intentional blurring.

CS3500 January 20, 2005
 80

View Volume Specification

• View volume is specified by 6 planes:
left, right, top, bottom, near, far.

• All values are given in camera coordinates.

But, near, far values are positive.

• Needn’t be symmetric!

• Can image the world sector by sector using both

orthographic and perspective cameras.

CS3500 January 20, 2005

 81

Alternate Specification
• Symmetric view volumes: horizontal and vertical fields of
view θh, θv

• For symmetric perspective view volumes:

tan θ2h = . . .?
tan θ2v = . . .?

• Aspect ratio: top / right.

CS3500 January 20, 2005

 82

End of Class 12

CS3500 January 20, 2005

 83

Projections: Summary
Perspective Orthographic
         
x 1 0 0 0 X x 1 0 0 0 X
y 0 1 0 0  Y  y 0 1 0 0  Y 
         
 =  =
     
 z  0 0 0 0  Z   z  0 0 0 0  Z 
   
w 0 0 d1 1 1 w 0 0 d1 1 1

d<∞ d=∞

CS3500 January 24, 2005

 84

View Volume Specification

• View volume is specified by 6 planes:
left, right, top, bottom, near, far.

• All values are given in camera coordinates.

But, near, far values are positive.

• Needn’t be symmetric!

• Can image the world sector by sector using both

orthographic and perspective cameras.

CS3500 January 24, 2005

 85

Canonical View Volume

• Projection is not performed right away; instead, map the
view volume to a cube of fixed dimensions, called the
canonical view volume or a standard view volume

• This is achieved by the normalizing transformation.

• Why?
? Easier to eliminate objects outside the view volume.
? Orthographic & perspective aren’t different.
? The z-coordinates not thrown away. (Used later!)

CS3500 January 24, 2005

 86

Canonical View Volume: Dimensions

• PHIGS: −1 ≤ x ≤ 1,
− 1 ≤ y ≤ 1, 0 ≤ z ≤ −1.
• OpenGL: −1 ≤ x ≤ 1,
− 1 ≤ y ≤ 1, −1 ≤ z ≤ 1.
• Canonical view volume
is the Orthographic View
Volume (with scaling along
x and y).

CS3500 January 24, 2005

 87

Orthographic Normalizing Matrix

• Lengths (right - left), (top - bottom) and (far - near)
scaled to 2.

• Shift origin so as to range from -1 to +1.

• Matrix??

CS3500 January 24, 2005

 88

• Matrix:
 right+left

2
right−left 0 0 − right−left
 

 0 2
top−bottom 0 − top+bottom
top−bottom


2
 far+near


 0 0 far−near far−near


0 0 0 1

CS3500 January 24, 2005

 89

Perspective Normalizing Matrix

• More complicated than orthographic case, as a frustum
has to be mapped to a cube. Do it in steps.
• First, scale the horizontal and vertical extents so that the
vertical and horizontal fields of view are 90 degres.
• A scaling transformation with sx =??, sy =??, sz = 1
• View volume is almost right except for a uniform scale.
• Next, scale uniformly so that the far plane is at -1. We will
also have −1 ≤ x, y ≤ 1 at the far plane after this.
• sx = sy = sz =??
CS3500 January 24, 2005
 90

x = far
z = far

θh / 2
45

x=1
z = −1

z = −1 z = −1

z=k
z=0

CS3500 January 24, 2005

 91

• M1 with sx = cot θ2h , sy = cot θ2v , sz = 1

1
• M2 with sx = sy = sz = far
 
cot θh/2
 far 0 0 0
 
cot θv /2
0 0 0
 
 
• M 1 M2 = 
 far 

1


 0 0 far 0 


0 0 0 1
 

near
• The near plane is now at k = far
.
• View volume fits into the canonical view volume, but is
CS3500 January 24, 2005
 92

still a frustum! Need scaling proportional to |z| to convert

it into a cube.

• This needs a perspective transformation; all the rest have

been affine.

• To divide by the depth, make last row [0 0 −1 0].

• Simultaneously send z = −k plane to 0 and z = −1 plane

to −1.

• Scale z by 1 − k and add a translation to do this.

CS3500 January 24, 2005

 93

Perspective Normalizing Txform

• Matrix
 for this step: 
1 0 0 0
 
0 1 0 0 
 
0 0 1 k 
 
 1−k 1−k 

0 0 −1 0
• Where do (x, y, -k) and (x, y,
-1) get transformed to?
• Frustum becomes a cube!

CS3500 January 24, 2005

 94

Actual Projection
• We have already performed the perspective division.

• Projection involves simply dropping the z coordinate and

scaling x-y to the viewport.

• Why go through with the z-coordinates?

• The ordering is preserved along the depth dimension.

z values can be used for visibility determination.

CS3500 January 24, 2005

 95

Where is the Film?

• Turns out: It does not matter.

• A final scaling is in the viewport transformations.

• As long as the film is in front of the camera, we will see

an upright image.

• Can consider the near plane as the film plane.

CS3500 January 24, 2005

 96

3D Module: Recapitulation
• 3D Graphics additionally involves projecting the 3D world
to the 2D image plane of the camera.

• Compute the 3D world with respect to the camera. Or

compute the relative geometry first.

• This involves a series of rigid transfomrations. For

complex objects/environments, each object or its part is
described in its own coordinate system.

• Modelling places these different objects in the world

CS3500 January 24, 2005
 97

coordinate system. This could involve a hierarchy of

transforms for objects made up of complex parts.

• View Orientation computes the world in the VRC.

• Camera can be perspective or parallel (orthographic,

oblique). 6 planes give the view volume and defines the
camera.

• View Mapping involves mapping the world to a canonical

view volume, which is an orthographic view volume.
This Normalizing Transformation has different forms for
parallel and perspective cameras.
CS3500 January 24, 2005
 98

• Projection and Clipping are easy to perform in the

canonical view volume. An image with dimensions from
-1 to +1 results.

• Viewport tranformation is the final step, involving a 2D

scaling and translation to map to window coordinates that
can be used to address the frame buffer.

• Given a description of the 3D world primitives, project

each point to 2D to get 2D primitives. These can be scan-
converted using standard algorithms.

CS3500 January 24, 2005

 99

End of Class 13

CS3500 January 24, 2005