Range data archive

Vision List Digest: Article 3, Volume 9, Issue 8
From: flynn@pixel.cps.msu.edu (Patrick J. Flynn)
Post-Followup: submission@VISLIST.com

I've made 44 range images (obtained from our Technical Arts
scanner) available for anonymous ftp from:

styrofoam.cps.msu.edu (IP address 35.8.56.144)

in the pub/images directory.

Some images contain one object, some contain several, with various
amounts of occlusion.

Direct *specific* questions about the images to me (flynn@cps.msu.edu).
General questions about range sensing are best answered by reading
the surveys by Jarvis (PAMI '83), Nitzan (PAMI '88), or Besl
(in the `Advances in Machine Vision' book by J. Sanz, pub. by Springer).


Here is the text of the README file in the images directory.

This directory contains a bunch of range images produced by the
MSU Pattern Recognition and Image Processing Lab's Technical
Arts 100X scanner (aka `White scanner'). You are free to use
these images to test your algorithms. If the images are to appear
in a published article, please acknowledge the MSU PRIP Lab as
the source of the images (you don't have to mention my name, though).

File format: rather than deal with all the goofy standards out
there for images (and to preserve the floating-point representation),
these images are compressed ASCII text files. Beware: they expand by
about 10x when uncompressed. I recommend that you keep them
compressed to save disk space. Many of you will probably convert
these files to your own `local' image format anyway.

Each image file has a three-line header giving the number of rows and
columns. This is followed by four images. The first is the
so-called 'flag' image, where a pixel value of 1 means the corresponding
(x,y,z) values at that pixel are valid. If the flag value is zero, you
should ignore the (x,y,z) components for that pixel.

Following the flag image is the image of X-coordinates, the image
of Y-coordinates, and the image of Z-coordinates. All are
floating-point images. Our White scanner is configured so that
each stripe of range values occupies one column in the image. We
sweep the object under the stripe with an XY table to get an image.
So the X coordinate image is a linear ramp; the X value is taken
from the absolute position of the X stage in the XY table (we don't
do anything in the Y direction at present). The Y value depends
on the column number of the pixel, and the Z value is the measured
range (in our lab, Z is the height above a table).

You can use the 3D coordinates of each range pixel, or you can
throw away the X and Y images, and concern yourself with the Z-value
alone. Note that the `aspect ratio' of the image doesn't
have to be 1, although I try to keep it in the neighborhood of 1.

Availability: I will try to keep these images available on
styrofoam.cps.msu.edu (35.8.56.144) until I leave MSU this summer.
If my next job has machines with internet access and some disk space,
I'll put them there.

Remember to use binary mode when you transfer the images.


http://www.vislist.com