Workshop Organization

Workshops Sponsor: FGnet (IST-2000-26434) Face and Gesture Recognition Working group
Workshop Chairman: James L. Crowley, INP Grenoble
Data Preparation and Web Administration: Julien Letessier and Nicolas Gourier, INRIA Rhone-Alpes

Workshop Organising Committee

James L. Crowley, INP Grenoble
Dr. Tim Cootes, University of Manchester
Prof. Gerhard Rigoll, TUM - Technical University of Munich
Prof. Eric Granum, Aalborg University
Dr. Sebastien Marcel, IDIAP
Dr. Andreas Lanitis, Cyprus College, Nicosia
James Ferryman, Univ. of Reading

Proposed Program Committee

(to be confirmed)

Dr. Tim Cootes, University of Manchester
Prof. Gerhard Rigoll, TUM - Technical University of Munich
Prof. Eric Granum, Aalborg University
Dr. Sebastien Marcel, IDIAP
Dr. Andreas Lanitis, Cyprus College, Nicosia
James Ferryman, University of Reading
Thomas Moeslund, Aalborg University
Rainer Stiefelhagen, University of Karlsruhe
Rainer Herpers, University of Bonn
Daniela Hall, INRIA Rhone-Alpes
Francois Berard, Universite Joseph Fourier
David Cristinacce, Manchester University
Pong C Yuen, Hong Kong Baptist University
Trevor Darrel, MIT AI Lab

About FGnet

FGnet is the European working group on face and gesture recognition funded by the EC IST program. The Coordinator of FGnet is Tim Cootes, University of Manchester, UK. Each year, FGnet published a set of benchmark image sequences on some aspect of visual observation of human activity. Resarch teams from around the world are invited to test their algorithms on the data set.

A workshop is organized at a major conference to showcase the results. Previous FGnet Workshops have included such data sets (PETS'02 at ECCV'02, and PETS'03 at ICVS'03).

The theme for the 2004 FGnet workshop is pointing gestures.

Important Dates

August 22, 2004: Venue
July 15, 2004: Camera-ready version
June 15, 2004: Notification of acceptance
April 30, 2004: Submission deadline

Call for Papers

The FGnet network has prepared video images and sequences of people pointing at targets with their hands and faces. Data has been prepared by projecting a target on a wall using a steerable video projector. Subject were seated in a room at a distance of 3 to 4 meters from the wall and asked to point at the target with their right hand and with their face. Unsynchronized video sequences were recorded from four cameras placed at different positions around the subject. Still images were also obtained for each gesture. For each gesture a ground truth was automatically recorded in the form of the known target position. Participants will be invited to compete in recognizing or estimating the target position with three categories of data:

stereo image pairs of hand pointing;
monocular images sequences of hand pointing;
monocular images of face pointing.

Two sets of sequences and still image sets are published on the workshop web site. Ground truth data is provided for the first set, but withheld from the second. Participants are required to submit a data file (format to be specified) in which they provide the target identity for each image or sequence. An automatic program will be run to compare submitted results to ground truth for the second sequence.

For each image or sequence, participants are invited to determine:

the precision of target estimation;
the probabiity of failure.

Participants will be requested to describe the theoretical foundations of their algorithm as well as computational cost of their implementation.

A one day workshop will be held in which authors present their algorithms and results. The workshop organizers will present the results from the participants. Copies of papers will be provided to workshop participants.

Pointing'04