Copy Link
Add to Bookmark
Report
VISION-LIST Digest Volume 13 Issue 16
VISION-LIST Digest Wed Apr 06 10:41:28 PDT 94 Volume 13 : Issue 16
- ***** The Vision List host is TELEOS.COM *****
- Send submissions to Vision-List@TELEOS.COM
- Vision List Digest available via COMP.AI.VISION newsgroup
- If you don't have access to COMP.AI.VISION, request list
membership to Vision-List-Request@TELEOS.COM
- Access Vision List Archives via anonymous ftp to FTP.TELEOS.COM
Today's Topics:
Face images
Need pose estimation from line correspondences
Lectureship in Vision Group at Univ of Surrey, U.K.
CFP: IJCAI-95
Advance Program: IEEE Workshop on Visualization and Machine Vision
Preliminary Program: Workshop on Visual Behaviors (long)
AI/GI/VI '94 Workshops/Tutorials - 16 to 17 May 1994
COMPMED 94 FINAL SCHEDULE
----------------------------------------------------------------------
Date: Tue, 5 Apr 1994 08:04:39 GMT
From: ackerman@iam.unibe.ch (Bernard Achermann)
Subject: Face images
Keywords: computer vision, face recognition
Hello everybody,
Several days ago I asked for informations about data bases containing
face images. And I promised a summary of my 'investigations'. Ok, here
it is.
First of all: Thanks to everyone who answered to my posting !!!
I found the following data bases:
1) MIT face database
*****************
ftp: whitechapel.media.mit.edu
directories: /pub/eigenfaces
/pub/images
short description:
This database obviously has been established by M. Turk and
Pentland. It contains 2592 images of 16 persons. For each person
there are 27 images taken under different conditions (lighting,
scale, ...). And for each image there are six levels of a
gaussian pyramid.
The images are stored in the file faceimages.tar.Z.
Additionally there's an example to the Eigenface paper of Turk/
Pentland. It's stored in the files faceimages-sample.tar.Z and
face-recognition.tar.Z.
Until now this is the best database I found for face recognition
purposes.
2) Imagery of the VISION-LIST-ARCHIVE
**********************************
ftp: ftp.teleos.com
directories: /VISION-LIST-ARCHIVE/IMAGERY/FACES
short description:
This database is a subset of the MIT face database. It contains
68 images.
3) USENIX face database (main site)
ftp: ftp.uu.net
directories: /published/usenix/faces
short description:
This is a database containing bitmaps of faces in a special format
for programs that show an image of a particular Internet user (i.e.
certain versions of finger). There are several archives (see below)
containing such bitmaps of users. Each user is stored in separate
file which contains additional information (name, institution ...).
Until now I didn't yet take a closer look at these images, but it
seems very probable that these USENIX collections are rather
heterogeneous (quality, illumination, ...).
4) USENIX face database (Indiana)
ftp: cs.indiana.edu
directories: /pub/faces
short description:
see point 3)
5) USENIX face database (London)
ftp: doc.ic.ac.uk
directories: /faces
short description:
see point 3)
I'm still looking for other sources of faces images. If
somebody finds some other data bases, please mail it to
me for I can update this list. Corrections and suggestions
are always appreciated!
See you
* Bernard Achermann *
* *
* E-Mail: ackerman@iam.unibe.ch *
* Tel. : + 41 31 631 49 02 *
* Institut fuer Informatik und * Institute of Computer Science and *
* angewandte Mathematik * Applied Mathematics *
* Universitaet Bern * University of Bern *
* Laenggass-Strasse 51 * Laenggass-Strasse 51 *
* CH-3012 Bern * CH-3012 Bern (Switzerland) *
------------------------------
Date: Wed, 30 Mar 1994 14:25:59 -0800
From: kapu@cs.washington.edu
Subject: Need pose estimation from line correspondences
I'm writing software for object recognition, and I need a piece of
software for pose estimation. I currently use software that gives
me pose estimation using pairs of corresponding points (3d model
points and 2d image points).
The primitives that I would like to use are lines, as lines contain
more information than points. Would anyone know/have any software
that would compute pose using line correspondences?
Please reply with e-mail: kapu@cs.washington.edu
Thanks!
--Kari Pulli
------------------------------
Date: Wed, 30 Mar 1994 10:25:42 +0100 (BST)
From: J.Illingworth@ee.surrey.ac.uk
Subject: Lectureship in Vision Group at Univ of Surrey, U.K.
UNIVERSITY OF SURREY
====================
Department of Electronic and Electrical Engineering
Vision, Speech and Signal Processing Group
******************************************
The Department has a Lectureship vacancy for a high-quality, research active
candidate in the area of Computer Vision, Image Processing and Pattern
Recognition. The Department is a large grade-5 research rated Department,
(one of only 6 in Elec Eng).
The Vision, Speech and Signal Processing Group is one of the largest computer
vision groups in the U.K. (over 30 members including academic staff,
Research Fellows and Research Students). It is headed by
Professor Josef Kittler and has a large portfolio of projects funded both
nationally and internationally. Facilities are excellent and include several
specialist facilities such as an active stereo vision head, a range data
scanner and robotic facilities. The group runs an advanced MSc course entitled
``Signal Processing and Machine Intelligence''.
The successful candidate would be expected to teach in the general software
area.
The University of Surrey is situated 30 miles from London in Guildford.
Guildford is a historic town with many excellent modern
facilities: good shopping, extensive sports facilities and many cultural and
entertainment venues (cinemas, theatres, Philharmonic Orchestra etc).
There are excellent communications links by road and rail and the
town is situated about 30 minutes from both of Londons major international
airports (Heathrow and Gatwick). The University is on a campus site only ten
minutes walk from the town centre.
The appointment will be on National Lecturers scales in the range
13,601-25,107 GBP, dependent on age and experience.
The closing date for consideration is 22 April 1994.
Applications in the form of a curriculum vitae (2 copies) with the names and
addresses of 3 referees should be sent to the Personnel Office (CVC/mah),
University of Surrey, Guildford GU2 5XH.
Anyone requiring informal details may contact the senior group members:
Professor Josef Kittler (0483) 259294
Dr John Illingworth (0483) 259835
Dr Maria Petrou (0483) 259801
Dr. J. Illingworth, | Phone: (0483) 259835
V.S.S.P. Group, | Fax : (0483) 34139
Dept of Electronic and Electrical Eng, | Email: J.Illingworth@ee.surrey.ac.uk
University of Surrey, |
Guildford, |
Surrey GU2 5XH |
United Kingdom |
------------------------------
Date: Fri, 1 Apr 94 11:18:58 PST
From: Rick Skalsky <skalsky@aaai.org>
Subject: CFP: IJCAI-95
CALL FOR PARTICIPATION: IJCAI-95
IJCAI-95 will take place at the Palais de Congres, Montreal, August 20-25
1995.
The biennial IJCAI conferences are the major forums for the international
scientific exchange and presentation of AI research. The Conference Technical
Program will include workshops, tutorials, panels and invited talks, as well
as tracks for paper and videotape presentations.
PAPER TRACK: SUBMISSION REQUIREMENTS AND GUIDELINES
Topics of Interest
Submissions are invited on substantial, original, and previously unpublished
research in all aspects of AI, including, but not limited to:
* Architectures and languages for AI (e.g. parallel hardware and software for
building AI systems)
* Artistic, entertainment and multimedia applications.
* Automated reasoning (e.g. theorem proving, abduction, automatic
programming, search, context management and truth maintenance systems,
constraint satisfaction, satisfiability checking)
* Cognitive modeling (e.g. user models, memory models)
* Connectionist and PDP models
* Distributed AI, autonomous agents, multi-agent systems and real-time
issues.
* Intelligent teaching systems
* Knowledge Engineering and Principles of AI applications (e.g. for design,
manufacturing control, grand challenge applications)
* Knowledge representation (e.g. logics for knowledge, action, belief and
intention, nonmonotonic formalisms, complexity analysis, languages
and systems for representing knowledge)
* Learning, knowledge acquisition and case-based reasoning
* Logic programming (e.g. semantics, deductive databases, relationships to
AI knowledge representation)
* Natural language (e.g. syntax, semantics, discourse, speech recognition
and understanding, natural language front ends, generation systems,
information extraction and retrieval)
* Philosophical foundations
* Planning and reasoning about action (including the relation between
planning and control)
* Qualitative reasoning and naive physics (e.g. temporal and spatial
reasoning, model-based reasoning, diagnosis)
* Reasoning under uncertainty (including fuzzy logic and fuzzy control)
* Robotic and artificial life systems (e.g. unmanned vehicles,
vision/manipulation systems)
* Social, economic and legal implications
* Vision (e.g. color, shape, stereo, motion, object recognition, active
vision, model-based vision, vision architectures and hardware, biological
modeling).
Timetable
Submissions must be received by 6th January 1995. Submissions received after
that date will be returned unopened. Authors should note that ordinary mail
can sometimes be considerably delayed, especially over the new year period,
and should take this into account when timing their submissions. Notification
of receipt will be mailed to the first author (or designated author) soon
after receipt.
Notification of acceptance or rejection: successful authors will be notified
on or before 20th March 1995. Unsuccessful authors will be notified by 27th
March 1995. Notification will be sent to the first author (or designated
author).
Camera ready copies of the final versions of accepted papers must be received
by the publisher in the USA by 24th April 1995.
Note that at least one author of each accepted paper is required to attend
the conference to present the work.
General
Authors should submit six (6) copies of their papers in hard copy form. All
paper submissions should be to the following address. Electronic or fax
submissions cannot be accepted.
IJCAI-95 Paper Submissions,
American Association for Artificial Intelligence,
445, Burgess Drive,
Menlo Park, CA. 94025, USA.
(telephone (415) 328-3123, email ijcai@aaai.org).
Appearance and Length
Papers should be printed on 8.5'' x 11'' or A4 sized paper. They must be a
maximum of 15 pages long, each page having no more than 43 lines, lines being
at most 140mm long and with 12 point type. Title, abstract, figures and
references must be included within this length limit. Papers breaking these
rules will not be considered for presentation at the conference.
Letter quality print is required. (Normally, dot-matrix printout will be
unacceptable unless truly of letter quality. Exceptions will be made for
submissions from countries where high quality printers are not widely
available.)
Title Page
Each copy of the paper must include a title page, separate from the body of
the paper. This should contain:
* Title of the paper
* Full names, postal addresses, phone numbers, fax numbers and email
addresses (where these exist) of all authors. The first postal address
should be one that is suitable for delivery of items by courier service
* An abstract of 100-200 words
* A set of keywords giving the area/subarea of the paper and describing the
topic of the paper. This information, together with the title of the paper,
will be the main information used in allocating reviewers.
* The following declaration:
``This paper has not already been accepted by and is not currently under
review for a journal or another conference. Nor will it be submitted
for such during IJCAI's review period.''
Policy on Multiple Submissions
IJCAI will not accept any paper which, at the time of submission, is under
review for a journal or another conference. Authors are also expected not to
submit their papers elsewhere during IJCAI's review period. These
restrictions apply only to journals and conferences, not to workshops and
similar specialized presentations with a limited audience.
Review Criteria
Papers will be subject to peer review, but this review will not be ``blind''
(that is, the reviewers will be aware of the names of the authors). Selection
criteria include accuracy and originality of ideas, clarity and significance
of results and the quality of the presentation. The decision of the Program
Committee, taking into consideration the individual reviews, will be final
and cannot be appealed. Papers selected will be scheduled for presentation
and will be printed in the proceedings. Authors of accepted papers, or their
representatives, are expected to present their papers at the conference.
Distinguished Paper Awards
The Program Committee will distinguish one or more papers of exceptional
quality for special awards. This decision will in no way depend on whether
the authors choose to enhance their paper with a video presentation.
Other Calls
Calls for tutorial and workshop proposals and video presentations for
IJCAI-95 will be issued shortly.
For questions or comments, (415) 328-3123, email ijcai@aaai.org
------------------------------
Date: Tue, 5 Apr 1994 16:22:55 -0400
From: rao@watson.ibm.com
Subject: Advance Program: IEEE Workshop on Visualization and Machine Vision
IEEE WORKSHOP ON VISUALIZATION AND MACHINE VISION
The Westin Hotel, Seattle, Washington June 24, 1994
ADVANCE PROGRAM
``Nonlinear models for representation, compression and
visualization of fluid flow images and velocimetry data,''
Ralph Ford and Robin Strickland
Dept. of Electrical and Computer Engineering
University of Arizona, Tucson, USA
``Boundary segmentation by detection of corner, inflection
and transition points''
Kazuhide Sugimoto and Fumiaki Tomita
Real World Computing Partnership Electrotechnical Laboratory
Ibaraki, Japan
``Resolving the topological ambiguity in approximating the
isosurface of a scalar function''
Sergey V. Matveyev
Institute for High Energy Physics
Moscow, Russia
``Simulation and visualization of integrated sensory-motor systems''
Suresh B. Marapane and Mohan Trivedi
Electrical and Computer Engineering Department
The University of Tennessee, Knoxville, USA
``Magnetic Contour Tracing''
Carol Orange and Francs Groen
Faculty of Applied Physics Faculty of Math and Comp. Science
Technical University of Delft University of Amsterdam
The Netherlands
``Exploring feature detection techniques for time-varying volumetric data''
Zhifan Zhu and Robert Moorhead II
NSF Engineering Research Center for Computational Field Simulation
Mississippi State University, USA
``Feature Visualization''
Deborah Silver and Norman Zabusky
Laboratory for Visiometrics and Modeling
Rutgers University
Piscataway, USA
``From Visualization to Perceptual Organization''
Boon-Lock Yeo and Shih-Ping Liou
Dept. of Electrical Engineering Imaging Department
Princeton University Siemens Corporate Research Inc.
Princeton, USA Princeton, USA
``Recognition of human facial expressions using 2-dimensional
physical model,''
Katsuhiro Matsuno, Chil-Woo Lee and Saburo Tsuji
Laboratories of Image Information Science and Technology
Osaka, Japan
``A Markov random fields model for describing unhomogeneous
textures: generalized random stereograms,''
Milan Jovovic
Electrical and Computer Engineering Department
University of Miami, Coral Gables, USA
``Extracting spatio-temporal patterns from geoscience
datasets,''
Edmond Mesrobian, Richard Muntz, Jose Santos, Eddie Shek and William Cheng
Data Mining Laboratory
Computer Science Department
University of California, Los Angeles, USA
PANEL DISCUSSION: Future directions in Visualization and Machine Vision
Panelists:
Lambertus Hesselink, Stanford University
Timothy Newman, National Institute of Health
Ingrid Carlbom, DEC Research Labs
Behzad Kamgar-Parsi, Naval Research Labs
------------------------------
Date: Tue, 29 Mar 94 17:58:58 EST
From: iaprwks@horus.cs.virginia.edu
Subject: Preliminary Program: Workshop on Visual Behaviors (long)
IAPR / IEEE
Workshop on Visual Behaviors
The Westin Hotel --- Seattle, Washington
June 19, 1994
in conjunction with CVPR'94
Few vision researchers question the statement,
"Vision must be understood as a spatio-temporal process."
However, there is still considerable debate about the
fundamental nature of that process. One major approach
claims that vision (and any perceptual process) can only be
understood properly in the context of motivating actions or
tasks. A further pragmatic observation is that the
complexity concerns require that limited system resources be
allocated dynamically in processing only restricted spatial
and temporal regions (at restricted resolutions). The
purpose of this workshop is to foster debate and dialogue on
the fundamental issues involved in systems that tightly
couple perception and behavior to perform tasks. The
program will consist of the papers given below and
associated system demonstrations.
PROGRAM COMMITTEE:
Worthy Martin University of Virginia, General Chair
Dana Ballard University of Rochester
Ruzena Bajcsy University of Pennsylvania
Tom Olson University of Virginia
Mike Swain University of Chicago
John Tsotsos University of Toronto
Linda Shapiro University of Washington, Local Arrangements
===========================================
IAPR/IEEE
Workshop on Visual Behaviors
Registration Form - Register Today!
===========================================
To register, just send this form and a check payable in US dollars to:
Prof. W.N. Martin
Department of Computer Science
Thornton Hall
University of Virginia
Charlottesville, Virginia 22093-2442
The check should be made out to: Workshop on Visual Behaviors.
For information send email to martin@virginia.edu
or call (804) 982-2202 (sorry, no phone registrations).
****************************************************************************
please type or print
Name ________________________________________________________________
Last/Family First Middle Badge Name
Company ________________________________________________________________
Address/Mailstop __________________________________________________________
City/State/Zip/Country ___________________________________________________
Daytime Phone _____________________ FAX Number ___________________________
Email ________________________________
(special pre-workshop information will be sent via email)
IAPR or IEEE/CS
Membership Number _________________________________________________
(required for member discount)
Do you have any special needs? ____________________________________________
Workshop Registration Fee
Advance (until 5/20/94) Late (after 5/20/94)
Member $35 $50
Nonmember $50 $65
Total Enclosed (in U.S. dollars) $ __________
Payment must be enclosed. Please make checks payable to:
Workshop on Visual Behaviors
All payments must be in U.S. dollars, drawn on U.S. banks.
Method of Payment Accepted
Personal Check Company Check Traveler's Check
Registration fee includes workshop attendance and
one copy of the workshop proceedings.
***************************************************************************
=================
Hotel Information
=================
The Seattle Westin is located in downtown Seattle within easy walking
distance of many restaurants and tourist attractions.
The workshop is being held in conjunction with CVPR in the Seattle Westin.
Transportation to the Seattle Westin from the Seattle-Tacoma Airport can be
booked through Gray Line Coach at the airport. The current price is $12 for
roundtrip transportation. A taxi from the airport to the hotel will cost
approximately $28 one way.
The CVPR conference rate for rooms at the Seattle Westin hotel will be $96
for a single and $111 for a double, plus taxes. The current tax rate is
15.2%. To obtain the conference rates at the Seattle Westin, make your
reservation by Monday May 30 at 5:00 pm Pacific time and mention that you
are attending CVPR. Reservations made after this date will be accepted by
the hotel on a "space available" basis. All major credit cards are accepted
by the hotel.
To make your reservation contact the hotel directly:
The Westin Hotel
1900 Fifth Avenue
Seattle, WA 98101
Phone: (206) 728-1000
Fax: (206) 728-2007
############################################################
# #
# Papers to be Presented #
# #
# #
############################################################
Memory Use During Hand-Eye Coordination
Dana H. Ballard, Mary M. Hayhoe and Jeff B. Pelz
University of Rochester
Abstract:
The very limited capacity of short-term or working
memory is one of the most prominent features of human
cognition. Most studies have stressed delimiting the upper
bounds of this memory in memorization tasks rather than the
performance of everyday tasks. We designed a series of
experiments to test the use of short-term memory in the
course of a natural hand-eye task where subjects have the
freedom to choose their own task parameters. In this case
subjects choose not to operate at the maximum capacity of
short-term memory but instead seek to minimize its use. In
particular, reducing the instantaneous memory required to
perform the task can be done by serializing the task with
eye movements. These eye movements allow subjects to
postpone the gathering of task-relevant information until
just before it is required. The reluctance to use short-
term memory can be explained if such memory is expensive to
use with respect to the cost of the serializing strategy.
******-
Integration of Camera Motion Behaviours for Active Object Recognition
David Wilkes and John Tsotsos
University of Toronto
Abstract:
The concept of active object recognition, as introduced
in our previous work, relies on the use of image-data-driven
behaviours to drive the camera to special views. We review
the arguments for an active approach, then argue for our
particular choice of behaviours. We address the issue of
integrating the behaviours into a complete recognition
methodology, and demonstrate its effectiveness in moderately
cluttered environments. Finally, areas for further work are
identified.
******-
Vision in Man-Made Environments: Looking for Syrup in all the
Right Places
Daniel Fu and Michael Swain
University of Chicago
Abstract:
People are often able to act efficiently in places like
various grocery stores, libraries, and other man-made
domains. They do so because they have useful knowledge
about how these domains are organized. In this paper we
show that an everyday domain, a grocery store, exhibits
useful regularities an agent can use both to simplify visual
tasks and to perform efficiently. We do this by first
identifying the organizations principles around which
grocery stores are structured, identifying the visual
routines involved, and then showing how our current
implementation, SHOPPER, uses these principles and routines
to find items in GROCERY WORLD, a simulated grocery store.
******-
The Surface-Attribute Probe: A "Smart" Tool for 3-D Object Recognition
Y. Tan and H. Freeman
Rutgers University
Abstract:
This paper describes a new approach to the classical
machine vision problem of 3-D object recognition and pose
determination. The process begins by extracting contour
edges from a given patterned-light image of an unknown
object and postulates an initial object line structure.
This line structure is used to guide placement of a set of
small surface-attribute-determining windows, called
"surface-attribute probes" (SAPs), over the image to
determine whether the areas being examined show the presence
of one or more intersecting planar or curved regions. The
resulting information is used to refine the object feature
description, i.e., to determine all face intersections and
characterize each face. The process is repeated until no
further refinement in region segmentation and surface-
attribute determination appears feasible. The ultimate
objective is to obtain (1) an accurate description of the
object line structure, (2) the orientation of all planar
faces, and (3) a qualitative characterization of all curved
surfaces. The segmentation process is guided by recursively
accessing data from the object-model data base, in which the
permissible objects are stored in terms of indexed multiview
representations. Experimental results are provided, showing
the status of the work to date.
******-
Spherical Retinal Flow for a Fixating Observer
Inigo Thomas, Eero Simoncelli and Ruzena Bajcsy
University of Pennsylvania
Abstract:
When a human observer moves, the eye continually
fixates on targets in the world. Although fixation is a
common process in human vision, its role has not yet been
established for computational purposes. The main
contribution of this paper is to formalize the retinal flow
for a fixating observer. A further contribution - a
potentially more practical one - is to explore the role of
the periphery in predicting collision. Utilizing fixation
is expected to turn out to be especially fruitful in light
of recent advances in computer vision for constructing
active head/eye systems [?].
In this work we make the following assumptions: (i) the
observer moves with respect to the world and fixates on a
target; (ii) the world is rigid, with no independently
moving elements; and (iii) the possible rotation axes of the
eye lie on a plane (comparable to Listing's Plane).
Assumptions (ii) and (iii) make the problem of determining
retinal flow tractable.
We first define retinal flow for a 2D universe and then
extend it to the full 3D case; the flow in 2D turns out to
form a component of the flow in 3. The retinal flow in 3D
will be decomposed into longitudinal and latitudinal flow;
the behavior of longitudinal flow along the retinal
periphery will be further analyzed for interesting
properties. Finally the results of a simulated experiment
on retinal flow at the periphery will be presented.
******-
Integration and Control of Reactive Visual Processes
James L Crowley and Jean Marc Bedrune
IMAG-LIFIA
Abstract:
This paper describes a new approach to the integration
and control of continuously operating visual processes.
Visual processes are expressed as transformations which map
signals from virtual sensors into commands for robotic
devices. These transformations define reactive processes
which tightly couple perception and action. Such
transformations may be used to control robotic devices, and
may also be used to control the fixation and optical
parameters of an active binocular head and the visual
processes which interpret visual data.
Our approach takes inspiration from so-called
"behavioural" approaches to mobility and manipulation.
However, unlike most previous work, we define reactive
transformations at the level of virtual sensors and device
controllers. This permits a system to integrate a large
number of perceptual processes and to dynamically compose
sequences of such processes to perform visual tasks. The
transition between visual processes is mediated by signals
from a supervisory controller as well as signals obtained
from perception. This method offers the possibility of
constructing vision systems with large numbers of visual
abilities in a manner which is both salable and learnable.
After a review of related work in mobility and
manipulation, we adapt the reactive process framework to
computer vision. We define reactive visual processes which
map information from virtual sensors to device commands. We
discuss the selection and control of reactive visual
processes to accomplish visual tasks. We then illustrate
this approach with a system which detects and fixates on
moving objects and on human faces.
******-
Representational Frames in Dynamic Scene Annotation
Aaron F. Bobick
MIT Media Lab
Abstract:
One of the tenets of machine vision is that the choice
of representation is critical in describing a scene;
representations are selected that make important information
explicit and that allow domain constraints to be brought to
bear. For describing dynamics scenes, however, there may be
no single representation appropriate for analysis; the
imagery is changing and the choice of best representation
may vary over time. We present two examples of where the
selection of representation frame - the coordinate system in
which the representation describes the scene - determined by
the dynamics of the scene. In the first example, we use
multiple, concurrent representations to describe some
action. These representations are self-validating, meaning
that they can test for their own applicability. The final
description of the action is a composite where part of the
action is described using one system, and the remainder
using another, as determined by the validity measures. The
second example is a simple demonstration of how the
transformation of imagery into the natural coordinate system
of the domain makes the application of domain knowledge
simple. Though this has always been the case, it is
particularly important in video annotation where
transforming domain knowledge into the imagery is more
difficult since the reference frame of the imagery may be
constantly changing.
******-
Perception and Action in a Dynamic Three-Dimensional World
Frank Z. Brill III
University of Virginia
Abstract:
Part of the artificial intelligence community has
broken away from the traditional approach and founded the
sub-discipline known as reactive planning. Reactive systems
do not maintain a world model; instead, they perform actions
reflexively, based primarily on immediate sensory
information obtained directly from the environment. There
are, however, fundamental limits on the capabilities of
systems that do not retain some information about the world.
In particular, a representation of the local space, i.e.,
spatial memory, is required to efficiently perform search
and avoidance tasks in a realistic environment. Given that
it is intractable to maintain a completely current and
accurate model of a dynamic world, research will investigate
the extent to which a world model is necessary for
accomplishing particular tasks in a dynamic three-
dimensional world, and the form of that model. These issues
are addressed by augmenting a reactive agent with a minimal
retention of spatial information necessary to achieve
competence in specific search and avoidance tasks in a
dynamic three-dimensional environment. Visual routines
(top-down, goal-directed visual computations) will be used
to implement the perceptual strategy, which will produce
deictic (agent-centered, activity-oriented) representations
of the world. The principal research in this area will
address the application of visual routines to three-
dimensional environments, the introduction of spatial memory
to deictic representations, and architectural techniques for
building autonomous agents which support these concepts.
******-
A Novel Environment for Situated Vision and Behavior
Trevor Darrell, Pattie Maes, Bruce Blumberg, Alex P. Pentland
MIT Media Lab
Abstract:
In this paper we present a new environment for the
development of situated vision and behavior algorithms. We
demonstrate a system which allows wireless full-body
interaction between a human participant and a graphical
world inhabited by autonomous agents. An image of the
participant is composited together with the graphical world
and projected onto a large screen in from of the
participant. No goggles, gloves, or wires are needed for
interaction with the world: agents and objects in the
graphical world can be acted upon by the human participant
through the use of domain-specific computer vision
techniques that analyze the silhouette and gestures of the
person. The agents inhabiting the world are modeled as
autonomous behaving entities which have their own sensors
and goals and which can interpret the actions of the
participant and react to them in real-time. We have
demonstrated and tested our system with two prototypical
worlds and describe the results obtained and lessons
learned.
******-
Task and Environment-Sensitive Tracking
Peter N. Prokopowicz, Michael J. Swain and Roger E. Kahn
University of Chicago
Abstract:
In a mobile robot, visual tracking, like other visual
behaviors, takes place in a context that includes aspects of
the task, the object being tracked, and the background. In
this work, prior knowledge of those task and target
characteristics that either enable or hinder different
real-time image-tracking algorithms, together with run-time
evaluation of the robot's environment, are used to select an
algorithm appropriate to the context.
******-
Directing Attention to Onset and Offset of Image - Events for
Eye-Head Movement Control
Winky Y. K. Wai and John K. Tsotsos
University of Toronto
Abstract:
This paper proposes a model that investigates a new
avenue for attention control based on dynamic scenes. We
have derived a computational model to detect abrupt changes
and have examined how the most prominent change can be
determined. With such a model, we explore the possibility
of an attentional mechanism, in part guided by abrupt
changes, for gaze control.
The computational model is derived from the difference
of Gaussian (DOG) model and it examines the change in the
response of the DOG operator over time to determine if
changes have occurred. On and off-DOG operators are used to
detect "on" and "off" events respectively. The response of
these operators is examined over various temporal window
sizes so that changes at different rates can be found. The
most salient "on" and "off" events are determined from the
corresponding winner-take=all (WTA) network. The model has
been tested with image sequences which have changes caused
by brightness or motion and the results are satisfactory.
******-
An Active Visual Attention System to "Play Where's Waldo"
W. E. L. Grimson, A. Lakshmi Ratan, P. A. O'Donnell and G. Klanderman
MIT AI Lab
Abstract:
We present an attentive active vision system, which
integrates visual cues to attentively fixate candidate
regions in which to recognize a target object. The system
uses a combination of color and stereo cues to perform
figure/ground separation, yielding candidates regions in
which to focus attention. Within each image region, we use
stereo to extract features that lie within a narrow
disparity range about the fixation position. These features
are used as input to an Alignment recognition system. We
demonstrate significant reductions in the complexity of
recognition using such methods. Our system can successfully
scan a large room, fixating and recognizing specified
targets. We also argue that cues such as stereo can be used
for figure/ground separation without needing accurate camera
calibration.
******-
Color Object Tracking with Adaptive Modeling
Rolf Schuster
Technical University, Munich and Siemens
Abstract:
The color representation of an object viewed with a
color camera is influenced by many parameters (ambient
light, object movement, camera characteristics, etc.). In
that context we address the problem of tracking an object
over time using adaptive color and adaptive shape modeling.
We develop two color models (Ellipsoid- and Mixture
Density Model) to describe the color of an object in two-
and three-dimensional color histogram space. For shape
modeling we suggest a simple moment based modeling scheme.
Finally, we propose an object tracking algorithm that
compensates for object color variations by adaptively
changing both, color and shape models (adaptive modeling).
The procedure has been tested on image sequences of the
human hand with changing ambient light, object movement and
relatively crowded background. The results show that
adaptive modeling performs reliable and fast object
tracking.
******-
Using Multiple Cues for Controlling an Agile Camera Head
Claus Siggaard Andersen and Henrik Iskov Christensen
Aalborg University
Abstract:
The interest in camera heads has been ever increasing
through the past years, and more and more places have them.
The interest in agile sensor systems seems promising for
guiding the vision research in new directions, one of these
being how to direct cameras to obtain a given goal. At
Aalborg University, we have implemented the second and much
improved version of a camera head, which has lead to
research in how to control the system, using visual
information. We propose using a combination of multiple
cues for controlling the camera head, enabling it to cope
with less restricted scenarios than otherwise possible. By
integrating accommodation and disparity cues we are able to
control the vergence of the camera head separated from the
version mechanism, which allows for a simple and yet
reliable method for doing smooth pursuit and base shift, a
central problem of camera head control, and focus of
attention.
******-
Visual Servoing Using Image Motion Information
V. Sundareswaran, F. Chaumette and P. Bouthemy
IRISA/INRIA
Abstract:
Visual servoing is a framework for achieving the tight
coupling of a camera's movements and information from
images. We consider a typical visual servoing approach that
uses geometric information about image features for
controlling the position and attitude of a camera. We claim
that image motion information can be used as well. We
substantiate this claim by presenting two different
approaches to visual tasks that use motion information. The
first one uses motion data which manifests as positional
information; in particular, we use the focus of expansion.
The second one incorporates the parameters of the affine
motion model in the control equations. We illustrate both
these approaches by means of a task to align the optical
axis of the camera with the translational direction, and
present results of experiments done with a six DOF robot.
The contribution of this work is in showing that a tight
coupling between the camera behavior and image motion is
possible.
******-
A $1000 Active Stereo Vision System
Ian Horswill
MIT AI Lab
Abstract:
We describe a simple disparity-based vergence algorithm
running on an ultra-low-cost stereo vision system built from
off-the-shelf components, and present performance data on
the complete system. The system includes a 68020-class
processor, two low-resolution b/w cameras and a 4 degree of
freedom active head. The system runs at up to 16Hz, costs
less than $1000US, and weights less than half a kilogram,
consumes only 3W of power, and is suitable for mounting on a
low cost mobile robot.
We argue that sophisticated vision systems suitable for
mobile robotics are feasible, and that active head systems
need not be as complicated or expensive as was once thought.
******-
Vision-Based Behavior Acquisition For A Shooting Robot By Using
A Reinforcement Learning
Minoru Asada, Shoichi Noda, Sukoya Tawaratsumita and Koh Hosoda
Osaka University
Abstract:
We propose a method which acquires a purposive behavior
for a mobile robot to shoot a ball into the goal by using
the Q-learning, one of the reinforcement learning methods.
A mobile robot (an agent) does not need to know any
parameters of the 3_d environment or its
kinematics/dynamics. Information about the changes of the
environment is only the image captured from a single TV
camera mounted on the robot. Only two parameters, the
velocity and orientation are control commands to be learned.
Image positions of the ball and the goal are used as state
variables which evaluate the effect of an action taken
during the learning process. After the learning process,
the robot tries to carry a ball near the goal and to shoot
it. Both computer simulation and real robot experiments are
shown, and we discuss on the role of vision in the context
of the vision-based reinforcement learning.
******-
Localizing Un-Calibrated, Reactive Camera Motion in an Object
Centered Coordinate System
Claus Brondgaard Madsen and Henrik Iskov Christensen
Aalborg University
Abstract:
We present elements of an analytic framework for an
approach to determining the angle between legs of junctions
in a polyhedral domain. From geometrical analyses we define
a set of qualitative visual events induced by moving the
camera in a reactive manner. These events provide means of
determining the position of the camera in relation to an
object centered coordinate system. The sole assumption
being an ability to perform fixation, the approach is
completely free of calibration, matching and absolute
motion.
An expression is derived relating the apparent angle
(viewed angle) to the true angle and the view point
specified in an object centered coordinate system. We show
that this relation has saddle surface characteristics. We
then define two camera motion strategies based on the
direction of the bisecting line in the image. Resulting
general trajectories are computed. Using the trajectories
we analyze the rate-of-change in apparent angle and the
orientation of the bisecting line. From these analyses we
define a set of visual events and relate them to what
knowledge about view point can be deduced from them.
Experiments are presented to illustrate the effect of
the motion strategies and to exemplify positional knowledge
obtained from sequences of visual events.
------------------------------
Date: Thu, 31 Mar 94 14:16:10 -0500
From: Colin Archibald <colin@autpi.iitsg.nrc.ca>
Subject: AI/GI/VI '94 Workshops/Tutorials - 16 to 17 May 1994
[ I have edited this announcement to only topics related to Image
Processing and Computer Vision. Please email colin@autpi.iitsg.nrc.ca
to get complete information.
phil... ]
AI/GI/VI '94 Workshops/Tutorials - 16 to 17 May 1994
These workshops are associated with the Artificial Intelligence, Graphics
Interface and Vision Interface conferences to be help in Banff, Alberta,
Canada on May 18-20. For more information on these three conferences
contact:
Tony Marsland Ph: (403) 492-3971, Fax: (403) 492-1071
email: tony@cs.ualberta.ca
Wayne Davis Ph: (604) 494-9056, Fax: (403) 494-4033
email:davis@cs.ualberta.ca
Distributed AI -- Mon 16 May -- 8:30 to 5:00
[ description omitted ]
Machine Learning -- Tue 17 May -- 8:30 to 5:00
[ description omitted ]
Tutorial - Image Processing with Khoros -- Tue 17 May -- 8:30 to 12:00
Gregory W Donohoe, University of New Mexico
Khoros is a portable software application environment that is especially
well suited to image processing. Originally developed at the University
of New Mexico, Khoros is available free to end users. There are between
10,000 and 20,000 users worldwide. Khoros is easily extensible; about a
dozen toolboxes and 100 image processing routines have been contributed
by researchers in Europe, Asia and North and South America. Like Khoros
itself, these are available free over the Internet by anonymous ftp.
This workshop will present an overview of Khoros and illustrate how
applications are constructed.
Tutorial - Evolutionary Algorithms -- Tue 17 May -- 1:30 to 5:00
Gerhard Roth, National Research Council
Evolutionary algorithms (EA) are powerful methods for solving various
search and optimization problems. The EA field includes such diverse
approaches as genetic algorithms (GA), evolutionary programming (EP),
classifier systems (CFS) and genetic programming (GP). All of these
different approaches are based on principles from the theory of
evolution. We will discuss applications of each approach for practical
problems in such diverse areas as scheduling, image processing, and
engineering design. Some commonly available public domain EA software
packages will be surveyed.
For more information for Image Processing & Evolutionary
Algorithms Workshops contact: Colin Archibald
National Research Council of Canada, Montreal Road, Ottawa, Ontario K1A 0R6
Phone: 613-993-6580 Fax: 613-952-0215 email: archibald@iit.nrc.ca
*****************************************************************************
Workshop Registration Form
AI/GI/VI '94 - Banff, Alberta - 16 to 20 May 1994
AI/GI/VI '94 Workshops - 16 to 17 May 1994
Family Name______________________________ Given Name___________________________
Affiliation____________________________________________________________________
Address________________________________________________________________________
City________________________________Province/State_____________________________
Country______________________________________Postal Code_______________________
Phone________________Fax________________e-mail_________________________________
Select below the workshop(s) that you wish to attend. The fee for each
workshop is $50.00 (CDN) for regular attendees and $25.00 (CDN) for
students.
Distributed AI _______
Machine Learning _______
Image Processing with Khoros _______
Evolutionary Algorithms _______ Total _______
Payment can be made by cheque or money order (made out to: AI/GI/VI'94)
or credit card (MasterCard or Visa only). Payment can be made on site,
or in advance by mail, fax, or e-mail. After May 2, please register on
site.
Credit Card: VISA M/C (Circle one)
Credit Card Number___________________________________Expiry Date_______________
Name on Credit Card__________________________Signature_________________________
Mail completed form to:
AI/GI/VI '94 Workshops
c/o Peter van Beek
Department of Computing Science
University of Alberta
Edmonton, Alberta, Canada T6G 2H1
or fax to: 403-492-1071
or e-mail to: vanbeek@cs.ualberta.ca
------------------------------
Date: Tue, 5 Apr 1994 14:46:50 +0600
From: mwitten@chpc.utexas.edu
Subject: COMPMED 94 FINAL SCHEDULE
FINAL PROGRAM ANNOUNCEMENT
FIRST WORLD CONGRESS ON COMPUTATIONAL MEDICINE AND PUBLIC HEALTH
24-28 April 1994
Hyatt on the Lake, Austin, Texas
The final program for the First World Congress On Computational
Medicine and Public Health has now been set. Over 200 speakers
will be presenting work in a variety of applications areas
related to medicine and public health. Registration is still
open for attendees. Registration details and/or a copy of the
schedule at a glance, schedule-in-detail may be requested by
sending an email request to
compmed94@chpc.utexas.edu
or by calling
512-471-2472
or by faxing
512-471-2445
There is no ftp form of the conference schedule due to the
size of the file. We will be happy to fax/send a copy to anyone
who requests it. The conference proceedings will appear as
a series of volumes published by World Scientific. If you are
interested in possibly submitting a paper for the proceedings,
please contact
mwitten@chpc.utexas.edu
or call
512-471-2457
The overwhelming response to this congress has already
justified having a second world congress in the future. The tentative
schedule is to have in in 3 years. If you are interested in
participating at the 2nd World Congress On Computational Medicine
and Public Health, please contact
Dr. Matthew Witten
Congress Chair
mwitten@chpc.utexas.edu
------------------------------
End of VISION-LIST digest 13.16
************************
