i
i
i
i
i
i
i
i
8.3. A Brief Look at Some Advanced Ideas in Computer Vision 209
Take the projected image of a scene with two cameras. Use image
processing software to identify key points in the picture from im-
age 1. Apply F to these points in turn to find the epipolar lines
in image 2 for each of these points. Now use the image process-
ing software to search along the epipolar lines and identify the
same point. With the location in both projections known and
the locations and orientations of t he cameras known, it is sim-
ply down to a bit of coordinate geometry to obtain the world
coordinate of the key points.
But there is more: if we can range find, we can do some very interesting
things in a virtual world. For example, we can have virtual elements in a
scene appear to interact with real elements in a scene viewed from two video
cameras. The virtual objects can appear to move in front of or behind the real
objects.
Let’s return to the question of how to specify and determine F and how
to apply it. As usual, we need to do a bit of mathematics, but before we do,
let us recap some vector rules that we will need to apply:
• Rule 1: The cross product of a vector with itself is equal to zero. For
example, X
3
×X
3
= 0.
• Rule 2: The dot product of a vector with itself can be rewritten as
the transpose of the vector matrix multiplied by itself. For example,
X
3
· X
3
= X
T
3
X
3
.
• Rule 3: Extending Rule 1, we can write X
3
· (X
3
× X
3
) = 0.
• Rule 4: This is not so much of a rule as an observation. The cross
product of a vector with another vector that has been transformed by a
matrix can be rewritten as a vector transformed by a different matrix.
For example, X
3
× T X
3
= MX
3
.
Remembering these four rules and using Figure 8.22 as a guide, let X
3
=
[x, y, z, w]
T
represent the coordinates in P
3
of the image of a point in the first
camera. Using the same notation, the image of that same point in the second
camera is at X
3
= [x
, y
, z
, w]
T
, also in world coordinates. These positions
are with reference to the world coordinate systems of each camera. After
projection onto the image planes, these points are at X = [X , Y , W ]
T
and
X
= [X
, Y
, W ]
T
, respectively, and these P
2
coordinates are again relative
to their respective projection planes. (Note: X
3
and X represent the same