This preview shows page 1. Sign up to view the full content.
Unformatted text preview: ne Bi as
,1
A,1
i
Bi = 0 i 0 ,A1 di
Let 01 = 1 B2 and 02 = 2 B1 . Let xi ; yi T be the
third column of 0i . In 15 it is shown that the epipolar
y
lines in image Ii make an angle of i = arctan x
with the horizontal axis. To make the epipolar lines
horizontal, each image is rotated by an angle of ,i ,
using the matrix: 5.3 Recti ed View Interpolation i Although image interpolation is not always physically valid, interpolation of recti ed monotonic images
always produces valid inbetween views of a scene. In
light of Eq. 3, it su ces to show that recti ed image
^
interpolation preserves monotonicity. Let Vi be an in^1 and V2 . Denote the
^
terpolation of recti ed views V
^^
^
rst row of 1 , 2 , and i by 1 , 2 , and i respectively. Because epipolar lines are horizontal, i.e.,
^
^
parallel to the image xaxis, X1 and X2 are in the same
epipolar plane. Therefore, the position of a scene point
P along its epipolar line in I^j is given by j P. Let P
and Q be two scene points in the same epipolar plane
of V1 and V2 . Suppose that monotonicity is satis ed
^
^
for I1 and I2 so that 1 P 1 Q and 2 P 2 Q.
Then
i P , Q = 2 , i1 + i , 12 P , Q
= 2 , i1 P , Q
+ i , 12 P , Q
6
Since both terms on the right of Eq. 6 are strictly
positive for 1 i 2, it follows that i P i Q and
hence monotonicity is preserved.
This result indicates that linear interpolation of corresponding points in two recti ed basis images always i cosi
R, = ,sini
i sini
cosi To ensure that corresponding epipolar lines share
the same numbered scanline, we must vertically
scale I2 with respect to I1 . Accordingly, let
R, 0
B = R, 210211 where R,2 021 is the rst
row of R,2 02 . If B is not invertible, either the
two views have parallel optical axes or the reference
features are coplanar. In either case, choose instead
01
B = R,01 1 . It follows that
0 R,2 02B,1 = 001
0s0 If s 0 it means that the epipolar lines in I2 are
horizontal but reversed with respect to I1 . In this case,
I2 should be rotated 180 degrees. To understand why
5 ^
Therefore Ii is an orthographic view of the scene
^
^
with aspect ratio kXi k : 1. In general, 0 kXi k 1,
^ i k decreasing monotonically as jj increases.
with kX
^
^
In particular, if V1 and V2 are within 45 degrees of one
^
another then kXi k is strictly greater than 0:92. In this
case, the greatest possible distortion, an 8 horizontal
contraction, occurs when i = 1:5, corresponding to a
^
^
view halfway between I1 and I2 .
If is known, this distortion can be avoided alto^
^
gether by scaling the rows of Ii by 1=kXi k. Although cannot be determined from the two basis images alone,
any third view of the scene is su cient to uniquely determine 16 . produces a valid view, assuming that monotonicity
holds for the basis images. The signi cance of this
result is that
1. Valid views can be produced by simple 2D image operations, without knowledge of either scene
structure or camera geometry, and
2. Morphing techniques based on geometric image
interpolation will produce physicallyvalid intermediate images if the basis...
View
Full
Document
This note was uploaded on 06/13/2011 for the course CAP 6938 taught by Professor Staff during the Spring '08 term at University of Central Florida.
 Spring '08
 Staff

Click to edit the document details