Copy Link
Add to Bookmark
Report
VISION-LIST Digest 1988 05 20
Vision-List Digest Fri May 20 16:06:02 PDT 1988
- Send submissions to Vision-List@ADS.COM
- Send requests for list membership to Vision-List-Request@ADS.COM
Today's Topics:
Video digitizers for PCs - any experiences?
Motion Workshop
Data Cube information request
(Computer controlled video lens) & (Calibration and cameras)
----------------------------------------------------------------------
Date: 7 May 88 21:38:55 GMT
From: steve@cs.hw.ac.uk (Steven Salvini)
Subject: Video digitizers for PCs - any experiences?
Keywords: video digitizer image processing frame-grabber
Organization: Computer Science, Heriot-Watt U., Scotland
Has anyone any experience with low cost (300 pounds sterling MAXIMUM!)
video digitizers for the PC? I am about to buy one and would value any
pointers...
At present all I've heard of are -
(1) The Electric Studio Video Digitiser - ^ 150:00 - no more info!
(2) Stem Computing - ^ 280:00 - 15 seconds/frame!
(3) "A firm in Livingston" ???name??? - ^ 120:00 - up to 5
frames/second
Please reply by e-mail and I'll summarise to the net.
Steve.
Steven Salvini JANET : steve@uk.ac.hw.cs
Department of Computer Science UUCP : ..!ukc!cs.hw.ac.uk!steve
Heriot-Watt University ARPA : steve@cs.hw.ac.uk
EDINBURGH EH1 3HJ Scotland Phone#: (+44) 31 225 6465 (Ext. 538)
------------------------------
Date: Tue, 17 May 88 16:13:06 EDT
From: jain@caen.engin.umich.edu (Ramesh Jain)
Subject: Motion Workshop
IEEE WORKSHOP ON VISUAL MOTION
Irvine, California
March 20-22, 1989
The analysis of visual motion is attracting increasing attention
from researchers in Computer Vision, Visual Perception and Artificial
Intelligence. This workshop will bring together researchers from this
broad range of disciplines to discuss current work on the
representation and analysis of motion in image sequences. Sessions
will be devoted to each of several aspects of motion research. The
number of presentations will be limited, with the aim of increasing
discussions. Papers are invited on all aspects of the analysis of
human and machine vision, including:
* Motion detection mechanisms
* Optical flow and motion correspondence
* Structure from Motion
* Event recognition and representation
* Temporal planning and inferences
* Control structures for dynamic scene analysis
* Uncertainty in dynamic scene analysis
* Applications in Navigation, object manipulation and recognition
Authors are encouraged to present new computational methods with
experimental results, theoretical results that offer significant new
insights into problem solutions, or to relate experimental observations
on human visual processing to the underlying computational strategies
used.
SUBMISSION OF PAPERS:
Submit three copies of the paper to either Ellen Hildreth or Ramesh
Jain on or before July 15, 1988. The papers should not exceed 25
double spaced pages. Authors will be notified of the decisions by
October 15, 1988. Final papers will be required by the Computer
Society of the IEEE on November 25, 1988.
General Chairman: Brian G. Schunck
Program Committee: Ted Adelson
John Aloimonos
Rama Chellappa
Allen Jepson
H.-H. Nagel
Ken Nakayama
Bill Thompson
Allen Waxman
Program Chairmen: Ellen Hildreth
Artificial Intelligence Laboratory
545 Technology Square
Cambridge, MA 02139
Ramesh Jain
Electrical Engineering and Computer Science
The University of Michigan
Ann Arbor, MI 48109-2122
Local Arrangements: M.L. Braunstein
D.D. Hoffman
------------------------------
Date: Fri, 20 May 88 17:37:18 EDT
From: Marc Bastuscheck <bastusch@acf8.NYU.EDU>
Subject: Data Cube information request
We are considering the purchase of some of the Datacube Max Video
VME-bus image processing boards for use in a Sun environment. We
currently have a VICOM-1, and would like the new facility to support
VICOM-like commands, support vision studies, and support real-time
robotic applications.
Our impression is that the Max Video series has plenty of
capabilities, but is difficult to program, and can get relatively
expensive if you load in a full complement of boards.
We would welcome comments, either public or private, on the following:
What sorts of real-time or near real-time projects have been
successfully performed using Datacube boards?
What is the longest "pipe" process, and how has that worked?
How accurate is the arithmetic? Do you routinely work with
16 bit images? Can one perform convolutions with 12 or 16
bit kernel elements to accumulate 16 bit precision?
Are there software packages available that give a VICOM-like
command set?
Send replies to "marc@acf8.nyu.edu". Thanks.
Marc Bastuscheck
[ Actually, because these questions are of great interest to many of
us on this List, please respond directly to VISION-LIST@ADS.COM.
I would like to see more of these types of questions and discussions.
-- pk ]
------------------------------
Date: Fri, 20 May 88 17:07:32 EDT
From: pkahn@ads.com (Phil Kahn)
Subject: (Computer controlled video lens) & (Calibration and cameras)
I have been checking into hardware to support active visual sensing
(e.g., as discussed by Gibson, the "active vision" school, etc.), and
the following questions have come up (any help or comments are
appreciated):
COMPUTER CONTROLLED VIDEO LENS:
I have been seeking an off-the-shelf video lens which is computer
controllable, but have not had much luck. Such a system would allow
the parameters of the lens (e.g., aperture, focus, zoom) to be
controlled via a computer interface (e.g., RS232 or plug-in board).
Such a system would allow the computer to specify the abolute position
of the optics (within some measureable error range). For example, I
would be able to have a program set the optics to focus maximally on
objects 20 feet away with an aperture of 6mm; as the object moves
closer to the camera, the computer could then shorten the focal length
appropriately.
The difficulty I have found is that motorized lenses (e.g.,
from Cosmicar (800)445-4233, Universe Kogaku (America) Inc
(516)671-8500, Computar (via Westech) (408)735-8687) do not contain
internal position sensors. These lenses support relative position
commands (e.g., "reduce the focal length"). Because their primary use
is for automatic 35MM cameras (which seek to obtain the "sharpest"
image) the control of lens parameters usually optimizes some
image-based contrast (for focus) or intensity (for aperture) metric.
A system with accurate computer control would require closed loop
control (hence the need for an in-the-lens position sensor) or a very
accurate open loop motor system (e.g., as obtained with
microsteppers).
Here's the question. Are there such computer controlled
lenses on the market for realistic cost? Alternatively, are there
specific firms you know which can fabricate such a beasty for
reasonable cost?
CALIBRATION AND CAMERAS:
Let's assume that such a computer controlled lens was available.
How would calibration then be accomplished? This includes geometric,
transduction, etc. calibration.
I've read a few papers for calibrating a camera with fixed
parameters (e.g., Tsai in IEEE J. Robotics and Automation, Aug. 1987;
Grosky and Tamburino, ICCV87), but it is not apparent to me how these
approaches can be used for a camera whose parameters can change and
result in known repeatability errors. I suppose that the camera could
be calibrated for a discrete enumeration of camera parameters (i.e.,
fix the number of optical parameter positions and calibrate for each
combination); greater accuracy could then be obtained by interpolating
calibration between these discrete parameter positions. Has anyone
written a paper on this?
Also, the calibration procedure is quite involved. In
particular, how can someone without an extensive background in optics
calibrate their camera? Is there a stock calibration station
configuration and software for performing the calibration? Are there
commercial services which can calibrate a camera? I know that Steve
Shafer's lab at CMU has addressed these problems...
The only constraint we have for a camera is that it: is small
in size, high in resolution (~512^2), use a chip sensor, and not have
autogain control. I have seen cameras which advertise square pixels
(e.g., VIDEK MEGAPLUS (800)44-KODAK); what are the issues here? Why
are square pixels better? Is it to better ensure symmetry under more
acute viewing angles? And what about sensor measurements? For
example, what cameras have had published accounts of their imaging
properties (e.g., as in the paper by Walton in Proc. Intern. Tech.
Symp. on Optical and Opto. Applied Sci. & Eng., vol. 693, Aug. 1986)?
I have been leaning towards the Pulnix TM540R, ~$1400, VERY small
(1.25" x 1.5" x 1.25") or the Pulnix TM-840, ~1900, small (46mm x 45mm
x 135mm), nearly square pixels (11.5x13.5), hi res (800 X 490 pixels),
with optional shutter and variable integration features. Yet, this
choice was based solely upon their promo sheets. The VIDEK seems good
(1320x1035 square pixels with no space between them and a built in
A/D!), though I don't know much more than that. Any comments? Other
cameras?
Whew.
------------------------------
End of VISION-LIST
********************
