home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
vis-ftp.cs.umass.edu
/
vis-ftp.cs.umass.edu.tar
/
vis-ftp.cs.umass.edu
/
pub
/
Motion
/
uil_stereo_motion
/
report
< prev
next >
Wrap
Text File
|
1990-11-05
|
12KB
|
458 lines
..........................................................................
. .
. This file can be printed out by the command "eqn | ptroff -me" .
..........................................................................
.ll 6.5i
.po 1.0i
.nr pp 12
.nr sp 12
.ps 12
.sz 12
.hy 14
.fo ''%''
.pn 1
.af % 1
.
.EQ
delim $$
define abs '| back 40 |'
define =del 'down 30 = back 60 {up 25 DELTA}'
define @ 'fat bold'
.gsize 11
.EN
.nf
.
$" "$
.sp |2.5i
.ce 2
\fBA SEQUENCE OF STEREO IMAGE DATA OF A MOVING VEHICLE\fP
.sp 0.2i
\fBIN AN OUTDOOR SCENE\fP
.sp 0.5i
.ce
Yuncai Liu and Thomas S. Huang
.sp 4.0i
.ce 5
Beckman Institute
University of Illinois at Urbana-Champaign
405 North Mathews Street
Urbana, IL 61801, U.S.A.
.sp 0.3i
August 10, 1990
.
.
.
.bp
.ce 2
\fBA SEQUENCE OF STEREO IMAGE DATA FROM A MOVING VEHICLE\fP
.sp 0.2i
\fBIN AN OUTDOOR SCENE\fP
.sp 2
.ce
Yuncai Liu and Thomas S. Huang
.sp 2
.ce 4
Beckman Institute
University of Illinois at Urbana-Champaign
405 North Mathews Street
Urbana, IL 61801, U.S.A.
.sp 3
.ls 2
.fi
.br
\fBI. Introduction\fP
.pp
A digital stereo image sequence data base has been created under the joint
effort of the University of Illinois at Urbana-Champaign, Purdue University
(Photogrammetry Group), and the U. S. Army Engineer Topographic Laboratory
(AI Center). Part of this data base, specifically three pairs of stereo
images taken at consecutive time instants (time instants 13, 14, and 15
from a long sequence) is now made available to the research community. More
will be available later. We hope that different algorithms will be
applied to these images and the results compared.
.pp
Two stationary cameras were used to take images of a moving vehicle in
an outdoor scene. The problem is to determine the 3D motion of the vehicle.
The image date can be used to test motion estimation algorithms based on
either stereo or monocular image sequences. In the latter case, only one
side of the sequence (from the left or the right camera only) is to be used.
Approximate ground truth for the vehicle motion has been established; however,
we choose not to reveal it to the public at large at this time.
.pp
Each of the six images has a size of 750 $times$ 750 pixels with 8 bits/pixel.
The image data are available on magnetic tape, and can be extracted by the
command \fItar xvf/dev/rst8 stereo\fP. In this report, we give formulas for
coverting the image point coordinates from pixels to the actual metric
distances on the image plane in the camera; and formulas for correcting
geometric distortions due to the camera lens.
.sp 2
.br
\fBII. Camera Setup\fP
.pp
The original stereo images were taken by two fixed focal length
A.M.I./Bronica SQ-AM metric 70 mm camera with 40 mm nominal focal length.
The cameras are parallelly mounted on the ground with a base line about
three meters. The optical axes of the cameras are in the
\fBnegative z-direction\fP of the ground coordinate system,
as shown in Figure 1.
.(z
.sp 4.0i
.ce
Figure 1. Relationship between the ground coordinate system and the cameras
.)z
If $(X',~Y')$ are the image coordinates of a point in the image plane of
a camera after geometric distortion correction, and $(x,~y,~z)$ are the
coordinates of the point in the ground coordinate system, the two are related
by
.EQ
left [~
pile {
{X' ~-~ X sub p} above {Y' ~-~ Y sub p} above {-~f} }
~ right ]
~=~ s @ M ~
left [~
pile {
{x ~-~ x sub c} above {y ~-~ y sub c} above {z ~-~ z sub c} }
~ right ]
.EN
.sp 0.5
.br
where $(X sub p , ~Y sub p )$ is the principal point of the camera and $f$
is the focal length, whose values for the left and the right cameras can be
found in Appendices A and B; $s$ is a scale factor; $@ M$ is a rotation
matrix with
.EQ
@ M ~=~
left [
pile {{roman "cos"~ kappa} above {- roman "sin"~ kappa} above 0}~~~
pile {{roman "sin"~ kappa} above {roman "cos"~ kappa} above 0}~~~
pile {0 above 0 above 1} right ]~
left [
pile {{roman "cos"~ phi} above 0 above {roman "sin"~ phi}}~~~
pile {0 above 1 above 0}~~~
pile {{- roman "sin"~ phi} above 0 above {roman "cos"~ phi}} right ]~
left [ ~
pile {1 above 0 above 0}~~~
pile {0 above {roman "cos"~ omega} above {- roman "sin"~ omega}}~~~
pile {0 above {roman "sin"~ omega} above {roman "cos"~ omega}} right ]
.EN
.sp 0.5
.br
where $omega$, $phi$ and $kappa$ are rotation angles; $(x sub c , ~y sub c ,~
z sub c )$ is the camera center position in the ground coordinate system. The
values of $(x sub c , ~y sub c , ~z sub c )$ and $omega$, $phi$ and $kappa$
are given in Appendix C.
.sp 2
\fBIII. Image Data Reduction\fP
.pp
Each image was digitized on a Perkin-Elmer scanning microdensitometer to
4096 $times$ 4096 pixels with 16 bits/pixel. The images are first subsampled
by a factor of 2, resulting in images of size 2048 $times$ 2048 pixels. Then,
subimages are cut out by a window of size 750 $times$ 750 pixels. See Figure 2.
.(z
.sp 4.0i
.ce
Figure 2. A subimage is cut out by a 750 $times$ 750 window
.sp
.)z
The coordinates of the upper-left corner of the window are (675, 415) in the
2048 $times$ 2048 images. If $(i, ~j)$ are the row and column numbers of a
point counting from top and left, respectively, in a 750 $times$ 750 image,
the coordinates of the point in the image plane of the camera should be
.EQ
X ~=~ [2~( Y sub cut ~+~ j ) ~-~ Y sub 0 ]~h
.EN
.
.EQ
Y ~=~ [ X sub 0 ~-~ 2~( X sub cut ~+~ i )]~w
.EN
.sp 0.5
.br
where $(X sub cut , ~ Y sub cut )$ are the coordinates of the upper-left
corner of the 750 $times$ 750 window with the values (675, 415) in the
2048 $times$ 2048 images; $(X sub 0 ,~ Y sub 0 )$ are the coordinates of
the image center of the corresponding 4096 $times$ 4096 images, whose value
will be given in Appendix D for each image; $h$ and $w$ are the height and
the width of a pixel in the original 4096 $times$ 4096 images: $h$ = $w$ =
$13 times 10 sup -3$ mm. The intensity range of the images is reduced to
8 bits.
.sp 2
.br
\fBIV. Image Data Correction\fP
.pp
The original images have inherent geometrical distortions. Two major
ones are the lens distortion due to the lens of a camera and the film
distortion due to the instability of film bases. Of the two
distortions, lens distortion has a dominant effect.
The film distortion is negligible. Therefore, only the
lens distortion is considered for geometric distortion correction. The
formulas used for lens distortion correction are
.EQ
X' ~=~ X ~+~ (X - X sub p )~(k sub 1 r sup 2 ~+~ k sub 2 r sup 4
~+~ k sub 3 r sup 6 )
~+~ 2 p sub 1 ~[ r sup 2 ~+~ 2 (X ~-~ X sub p ) sup 2 ]
~+~ p sub 2 (X ~-~ X sub p ) ( Y - Y sub p )
.EN
.
.EQ
Y' ~=~ Y ~+~ (Y - Y sub p )~(k sub 1 r sup 2 ~+~ k sub 2 r sup 4
~+~ k sub 3 r sup 6 )
~+~ 2 p sub 1 (X ~-~ X sub p ) ( Y - Y sub p )
~+~ p sub 2 ~[ r sup 2 ~+~ 2 (Y ~-~ Y sub p ) sup 2 ]
.EN
.sp 0.5
where
.ip
$(X',~Y')$ are the corrected image coordinates
.ip
$(X,~Y)$ are the image coordinates before correction.
.ip
$(X sub p ,~ Y sub p )$ are the coordinates of the calibrated principal
point of the lens.
.ip
$r ~=~ sqrt {(X ~-~ X sub p ) sup 2 ~+~ (Y ~-~ Y sub p ) sup 2 }$,
the distance from point $(X,Y)$ to the principal point.
.ip
$k sub 1$, $k sub 2$, $k sub 3$ are the radial distortion coefficients
of the lens.
.ip
$p sub 1$, $p sub 2$ are the decentering distortion coefficients of the
lens.
.lp
.br
All the calibration parameters of the lenses are given in Appendices A and B.
.sp 2
.br
\fBAcknowledgement\fP
.sp 0.5
.pp
The presons responsible for the creation of the image sequence data base are:
.br
University of Illinois: T. S. Huang, S. D. Blosteine, Y. C. Liu, M. K. Leung.
.br
U. S. Army Engineer Topographic Laboratory: M. McDonald, W. Seemuller, A.
Werkheiser.
.br
Purdue University: E. M. Mikhail, F. C. Paderes.
.pp
At University of Illinois, the work has been supported by National Science
Foundation Grant IRI-89-08255 and Grant IRI-89-02728, and the State of Illinois
Department of Commerce and Community Affairs Grant SCCA 90-82140.
.
.
.
.bp
.
.ce 2
\fBAPPENDIX A\fP
\fBCalibration Report for Left Camera\fP
.sp
.2c
.br
Camera Type: A.M.I./Bronica SQ-AM
.br
Lens Type: Zenzanon-S
.bc
.br
Nom. focal length: 40 mm
.br
Test focal setting: infinity
.1c
.sp 2
.br
1) Principal Distance:$~~~~~~~~~~$ f = 41.357 mm $+-$ 0.007 mm
.br
2)
Calibrated Principal Point:
.2c
.br
.ce
$X sub p$ = 0.035 mm $~~~~+-$0.005 mm
.
.br
.ce
$Y sub p$ = - 0.019 mm $~~~~+-$0.006 mm
.1c
.sp
.br
3) Calibrated Lens Distortion Parameters:
.
.ip
$k sub 1 ~=~ ~~5.32800E-05 ~~~~+- 4.10E-07$
.ip
$k sub 2 ~=~ - 4.12600E-08 ~~~~+- 5.30E-10$
.ip
$k sub 3 ~=~ ~~8.58900E-12 ~~~~+- 1.90E-13$
.ip
$p sub 1 ~=~ - 4.16400E-06 ~~~~+- 1.20E-06$
.ip
$p sub 2 ~=~ - 3.87400E-06 ~~~~+- 1.20E-06$
.
.
.
.bp
.
.ce 2
\fBAPPENDIX B\fP
\fBCalibration Report for Right Camera\fP
.sp
.2c
.br
Camera Type: A.M.I./Bronica SQ-AM
.br
Lens Type: Zenzanon-S
.bc
.br
Nom. focal length: 40 mm
.br
Test focal setting: infinity
.1c
.sp 2
.br
1) Principal Distance:$~~~~~~~~~~$ f = 41.337 mm $+-$ 0.007 mm
.br
2)
Calibrated Principal Point:
.2c
.br
.ce
$X sub p$ = 0.020 mm $~~~~+-$0.003 mm
.
.br
.ce
$Y sub p$ = 0.037 mm $~~~~+-$0.005 mm
.1c
.sp
.br
3) Calibrated Lens Distortion Parameters:
.
.ip
$k sub 1 ~=~ ~~4.99579E-05 ~~~~+- 5.17E-07$
.ip
$k sub 2 ~=~ - 3.62418E-08 ~~~~+- 8.74E-10$
.ip
$k sub 3 ~=~ ~~5.41550E-12 ~~~~+- 4.85E-13$
.ip
$p sub 1 ~=~ - 2.29110E-06 ~~~~+- 5.38E-07$
.ip
$p sub 2 ~=~ - 7.36123E-06 ~~~~+- 6.02E-07$
.
.
.
.bp
.
.ce 3
\fBAPPENDIX C\fP
.ls 1
\fBCAMERA POSITIONS AND ORIENTATIONS\fP
\fBMEASURED IN THE GROUND COORDINATE SYSTEM\fP
.ls 2
.sp 2
.br
1. Left Camera:
.sp 0.5
.br
Image frame 13:
.br
$(x sub c ,~ y sub c , ~ z sub c )$ = (35413.20, 1985.56, 68171.65)
.br
$( omega , ~ phi , ~ kappa )$ =
$(-~ 0 sup 0$ 21 ' 54.703",
$9 sup 0$ 3' 23.145",
$-~1 sup 0$ 1' 27.530")
.sp 0.5
.br
Image frame 14:
.br
$(x sub c ,~ y sub c , ~ z sub c )$ = (35413.94, 1989.10, 68175.55)
.br
$( omega , ~ phi , ~ kappa )$ =
$(- ~0 sup 0$ 22 ' 13.866",
$9 sup 0$ 3' 32.466",
$- ~1 sup 0$ 1' 19.863")
.sp 0.5
.br
Image frame 15:
.br
$(x sub c ,~ y sub c , ~ z sub c )$ = (35411.88, 1986.32, 68168.27)
.br
$( omega , ~ phi , ~ kappa )$ =
$(-~ 0 sup 0$ 21 ' 59.404",
$9 sup 0$ 3' 27.535",
$-~ 1 sup 0$ 1' 36.469")
.
.sp
.br
2. Right Camera:
.sp 0.5
.br
Image frame 13:
.br
$(x sub c ,~ y sub c , ~ z sub c )$ = (38418.24, 1979.32, 67881.57)
.br
$( omega , ~ phi , ~ kappa )$ =
$(-~ 0 sup 0$ 43 ' 58.416",
$9 sup 0$ 43' 52.854",
$- ~ 1 sup 0$ 7' 2.598")
.sp 0.5
.br
Image frame 14:
.br
$(x sub c ,~ y sub c , ~ z sub c )$ = (38417.79, 1977.91, 67879.94)
.br
$( omega , ~ phi , ~ kappa )$ =
$(-~ 0 sup 0$ 44 ' 2.496",
$9 sup 0$ 43' 54.974",
$-~ 1 sup 0$ 7' 16.905")
.sp 0.5
.bp
.br
Image frame 15:
.br
$(x sub c ,~ y sub c , ~ z sub c )$ = (38418.54, 1977.87, 67879.70)
.br
$( omega , ~ phi , ~ kappa )$ =
$(-~ 0 sup 0$ 43 ' 48.873",
$9 sup 0$ 43' 0.437",
$-~ 1 sup 0$ 7' 38.582")
.
.
.
.bp
.
.ce 4
\fBAPPENDIX D\fP
.ls 1
\fBPARAMETERS FOR THE IMAGE COORDINATE TRANSFORMATION\fP
\fBFROM THE DISCRETE SYSTEM\fP
\fBTO THE SYSTEM IN THE IMAGE PLANE OF A CAMERA\fP
.sp 2
.nf
.ls 2
.sp
.EQ
X ~=~ [2~( Y sub cut ~+~ j ) ~-~ Y sub 0 ]~h
.EN
.
.EQ
Y ~=~ [ X sub 0 ~-~ 2~( X sub cut ~+~ i )]~w
.EN
.sp 0.5
where
$(X sub cut ,~ Y sub cut )$ = (675, 415)
$h ~=~ 13 times 10 sup -3$ mm, $h ~=~ 13 times 10 sup -3$ mm
.sp
$(X sub 0 ,~ Y sub 0)$ are different for different images.
.sp
1. Left Camera:
Image frame 13: $(X sub 0 , ~Y sub 0 )$ = (2075, 2055)
Image frame 14: $(X sub 0 , ~Y sub 0 )$ = (2081, 2064)
Image frame 15: $(X sub 0 , ~Y sub 0 )$ = (2061, 2068)
.sp
2. Right Camera:
Image frame 13: $(X sub 0 , ~Y sub 0 )$ = (2071, 2055)
Image frame 14: $(X sub 0 , ~Y sub 0 )$ = (2082, 2052)
Image frame 15: $(X sub 0 , ~Y sub 0 )$ = (2070, 2057)
.