IMTC has been
working on a technology using video to extract images. The software
we have developed can perform two functions in its current state:
generation of panoramic images from controlled-panoramic video
and generation of a side-view image from driving along a street
at around 5-lOmph. We are planning to soon offer an Adobe Premiere
plugin that will produce these images automatically from the source
video. These images can be incoporated in most of the interactive
panoramic viewers available today.
Panoramic Image generation (QTVR)
For the first function (creating
an image from panoramic video), we have produced a number of
panoramic images from video by mounting
the camera rotated 90 degrees, so what is normally horizontal
is now vertical. We then rotated the camera about the base at a
speed
that would achieve square pixels. After digitizing the video,
we ran it through custom software to produce an image. The image
produced
yields a vertical resolution of 640 pixels and -- the number
of fields in the pan -- pixels in the horizontal (original video
digitized
at resolution = 640x480). Processing time was 4 times real time
or 4 sec per 1 sec of video. The pan start and finish overlap
to allow
for aligning and cropping of the image in an image manipulation
package to yield an image that is QTVR-ready. One test we
did
using a DV
camera with auto-exposure turned on yielded a continuous image
across 7 f-stops.
|
|
image from the Orthographic Mapper
SIGGRAPH ‘97
Presentation Slides
 |
Quicktime VR from atop the GCATT building
|
Orthogonal to Path of travel Image Creation (DriveBy) For the
second function, our main goal was to provide an easy method
of generating texture maps from a real environment where
photos would
not be feasible, namely, a 4-5 mile stretch of road for use in
a cycling simulator. Since this simulator was the sole source
of funding
for this research, we focused mainly on this application of the
technology. We mounted the camera on a car looking out perpendicular
to the path
of travel, again rotated 90 degrees so we have a larger vertical
resolution. We then drove down the road at about 5-10mph. After
digitizing the video and running it through the panoramic
setting for the software,
we were able to determine what other corrections needed to be
performed. Bumps in the road and inconsistent driving speeds resulted
in distortion
of the image. The software in its current state performs both
integer pixel correlation and subpixel correlation (up to the user)
to
correct for bumps in the road. Also built into the software,
is a function
to correct for varying speeds,so that all objects at the same
distance from the camera are proportional to each other. We have
further
developed the technology to produce two images that could
be used for stereo
(both image and audio) panoramic viewing.
|
source image for the following Quicktime movies
 |
 |
 |
 |
The Quicktime VR on the left is a 2D panoramic image showing a
drive down, a u-turn, and back up the wrong- side of the street
in the Buckhead area of Atlanta at about 5-10mph. (file download
- 2M) The next 3 Quicktime movies (file downloads - 2.5M, 1.45M,
13M) are reconstructed “2.5D” drive-by,
with the camera at 0 degrees and at 45 degrees.
|
]
[
