3-D Transformations

Brian Romsek
Senior Student
Surveying Engineering Department
 Three-Dimensional Conformal
Coordinate Transformation
• Converting from one three-dimensional system to another,
while preserving the true shape.

• This type of coordinate transformation is essential in analytical

photogrammetry to transform arbitrary stereo model
coordinates to a ground or object space system.

• It is often used in Geodesy to convert GPS coordinates in

WGS84 to State Plane Coordinates. is
Z Axis

Y Ax

X Axis
 Applications of 3D Conformal
Coordinate Transformations
•Mobile mapping systems
•Relations between different coordinate frames
•Sensor frame
•Body frame
•Mapping frame
 Applications of 3D Conformal
Coordinate Transformations
•Homeland security
•E.G., facial pattern recognition
•Image processing
 3D Conformal Coordinate
Transformation
• Also known as the 7 Parameters transformation since it
involves:
• Three rotation angles omega (), phi (), and kappa ();
• Three translation parameters (TX, TY,TZ) and
• a scale factor, S
Kappa ()

Z-axis
(X,Y,Z)
xis
Y- a
X-a
xis
Omega
()
Phi ()
 Rotation angles Omega
In general form:
X 2 X 1  Y1 0  Z1 0
Y2 X 1 0  Y1 cos   Z1 sin 
Z 2 X 1 0  Y1 ( sin )  Z1 cos 

In matrix form:
X2  1 0 0   X1 
 Y    0 cos  sin    Y 
 2    1
 Z 2   0  sin  cos   Z1 

More concisely C2 M  C
 Rotation angles Phi
In general form:
X 3 X 2 cos   Y2 0  Z 2 ( sin )
Y3 X 2 0  Y2  Z 2 0
Z 3 X 2 sin   Y2 0  Z 2 cos 
In matrix form:
Kappa ()
 X 3   cos  0  sin  X2 
 Y   0  Y 

Z-axis
  3 1 0   2
 Z 3   sin  0 cos    Z 2 
Omega
More concisely C M C X-
ax
()
3  2 is
Phi ()
 Rotation angles Kappa
In general form:
X' X 3 cos   Y3 sin   Z 3 0
Y' X 3  sin    Y3 cos   Z 3 0
Z' X 3 0  Y3 0  Z 3

In matrix form:
 X'  cos  sin  0  X3  Kappa ()
 Y'    sin  cos  0 Y 

Z-axis
     3
 Z'   0 0 1  Z 3 
Omega
More concisely C' M  C 3 X-
ax
()
is
Phi ()
 Combined Rotation Matrix
If we combine all the rotation matrices
 X'  X1   X1 
 Y' M  Y  M M M Y 
  G 1     1
 Z'   Z1   Z1 

 m11 m12 m13 

 
M M  M  M   m21 m22 m23 
m m32 m33 
 31
MG becomes, after multiplication
 cos  cos  cos  sin   sin  sin  cos  sin  sin   cos  sin  cos  
M G    cos  sin  cos  cos   sin  sin  sin  sin  cos   cos  sin  sin  
 sin   sin  cos  cos  cos  
COMPUTING ROTATION ANGLES

• If rotation matrix  m 32
tan  
known, rotation m 33
angles can be
computed as shown
on the right sin  m 31

 m 21
tan  
m11
 Properties of rotation matrix

• The rotation matrix is an orthogonal matrix,

which has the property that its inverse is equal
to its transpose, or 1 T
R R
• This can be used for inverse relationship
 Three-Dimensional Conformal
Coordinate Transformation
• Finally the 3D Conformal Transformation is derived
by multiplying the system by a scale factor s and adding
the translation factors TX, TY, and TZ.

C ' s M C  T
• Where:
 TX  X  m11 m12 m13 
    M  m 
T  TY  C ' Y   21 m22 m23 
T  Z m m33 

is
m32

Z Axis
 Z     31

Y Ax
X Axis
BURSA-WOLF TRANSFORMATION
• Geodesy assumption – rotation angles small
– cos  = 1
– sin  =  (radians)
– Product of two sines = 0

• Rotation matrix R becomes:

 1 Rκ  R 
 
R   Rκ 1 Rω 
 R  Rω 1 

BURSA-WOLF TRANSFORMATION
• 3D similarity transformation

 f(x)  X   0 Rκ  R   x   x   TX 
 g(x)  Y    R 0

Rω   y   s  y    TY 
     κ
 h(x)  Z   R  Rω 0   z   z   TZ 

• Observation Equation:

V BΔ  f
 BURSA-WOLF TRANSFORMATION
• Coefficient matrix, B:

1 0 0 x1 0  z1 y1 
0 1 0 y1 z1 0 
 x1 

0 0 1 z1  y1 x1 0 
B  
1 0 0 x2 0  z2 y2 
      
 
 0 0 1 z n  yn xn 0 
• Vector of parameters, , and
discrepancy vector, f X
f  Y 
Δ T X R 
T
TY TZ s R R  Z 
 Three Dimensional Coordinates
Transformation
General polynomial approach: transformation is not
conformal
Xn a0  a1 x  a2 y  a3 z  a4 x 2  a5 y 2  a6 z 2  a7 xy  a8 yz
a9 zx  a10 xy 2  a11 x 2 y  a12 xz 2  

Yn b0  b1 x  b2 y  b3 z  b4 x 2  b5 y 2  b6 z 2  b7 xy  b8 yz
b9 zx  b10 xy 2  b11 x 2 y  b12 xz 2  

Zn c0  c1 x  c2 y  c3 z  c4 x 2  c5 y 2  c6 z 2  c7 xy  c8 yz
c9 zx  c10 xy 2  c11 x 2 y  c12 xz 2 
 Three Dimensional Coordinates
Transformation
Alternative that is conformal in the three planes

Xn  A0  A1 x  A2 y  A3 z  A5  x 2  y 2  z 2   0  aA7 zx  2 A6 xy  

Yn B0  A2 x  A1 y  A4 z  A6  x 2  y 2  z 2   2 A7 yz  0  2 A5 xy  

Zn C0  A3 x  A4 y  A1 z  A7  x 2  y 2  z 2   2 A6 yz  2 A5 zx  0  
 Three Dimensional Coordinates
Transformation
a1 x  a2 y  a3 z  a4
Xn 
Polynomial d1 x  d 2 y  d3 z  1
projective
transformation, 15
b1 x  b2 y  b3 z  b4
parameters Yn 
d1 x  d 2 y  d3 z  1

c1 x  c2 y  c3 z  c4
Zn 
d1 x  d 2 y  d3 z  1
 Testing – 4 Methods
• Bursa Wolf • Generalized
– Linear model – Bursa Wolf
assume small – Linear model –
rotation angles errors in both
– Best for satellite observations and
to global system model parameters
transformations – Useful
– Bazlov et al: transforming
determined PX classical to space-
90 to WGS 84 borne (Kashani,
parameters 2006)
 Testing – 4 Methods
• Polynomial • Expanded Full- Model
– 1st order – Photogrammetric
– Useful when approach
coordinate – Angles not considered
systems not small
uniform in – Non-linear: requires a
orientation or priori estimate of
scale parameters
– Rubber-sheeting
 Expanded Full-Model
• Employed method shown in
“Photogrammetric Guide” by Abertz &
Kreiling
• X, Y, Z coordinates translated to
relative values based in mean
coordinates
 3D
3D Transformations
Transformations Testing
Testing
• Data include a set of know control points, transformed from
WGS84 system to State Plane Coordinates.
 Test Results
Method RMSE
Photo Guide 0.048221

Bursa-Wolf 0.130928

Generalized Bursa-Wolf 0.128114

n c e Polynomial 1st Order 0.211124

f e re e
Re rianc
Va
T
V WV
0  RMSE  ( X  Xˆ ) 2  (Y  Yˆ ) 2  ( Z  Zˆ ) 2 / n
2

n u