Articles in Control Magazine

Download our regular contributions to Control Magazine.

Issue 30, March 2012
Issue 29, January 2012
Issue 28, December 2011
Issue 27, November 2011
Issue 26, August 2011
Issue 25, July 2011
Issue 24, April 2011
Issue 23, February 2011
Issue 22, January 2011
Issue 21, November 2010
Issue 20, October 2010
Issue 18, June 2010
Issue 17, April 2010
Issue 16, January 2010

Control International Edition July 2011
Control International Edition March 2011
Control International Edition April 2010
Control International Edition August 2010 

About GATE

GATE final publication 2012
Results from the GATE research project
a 75 page overview (pfd 4.7 Mb)

GATE Magazine 2010
a 36-page overview of the GATE project (pdf 5.3 Mb

Research themes:
Theme 1: Modeling the virtual World
Theme 2: Virtual characters
Theme 3: Interacting with the world
Theme 4: Learning with simulated worlds

Pilots:
Pilot Education Story Box
Pilot Education Carkit
Pilot Safety Crisis management
Pilot Healthcare Scottie
Pilot Healthcare Wiihabilitainment

Knowledge Transfer Projects:
Sound Design 
CIGA 
Agecis 
CycART 
VidART
Motion Controller
Compliance
Mobile Learning
Glengarry Glen Ross
CASSIB
EIS
Enriching Geo-Specific Terrain
Pedestrian and Vehicle Traffic Interactions
Semantic Building Blocks for Declarative Virtual World Creation 
Computer Animation for Social Signals and Interactive Behaviors

Address

Center for Advanced Gaming and Simulation
Department of Information and Computing Sciences
Utrecht University
P.O. Box 80089
3508 TB Utrecht
The Netherlands
Tel +31 30 2537088

Acknowledgement

 ICTRegie is a compact, independent organisation consisting of a Supervisory Board, an Advisory Council, a director and a bureau. The Minister of Economic Affairs, and the Minister of Education, Culture and Science bear the political responsibility for ICTRegie. The organisation is supported by the Netherlands Organisation for Scientific Research (NWO) and SenterNovem.

VidART

Knowledge Transfer Project
VidART - Video Analysis and Recognition Toolbox

A toolbox for automatic human pose and gesture recognition
Human behavior studies are complex and manual annotation of the desired behaviors is tedious. To automate the identification and analysis of human poses and gestures from videos, requirements must be set about the scene. Another prerequisite is a robust method to detect people in the scene. Noldus Information Technology BV has developed software to manually annotate and analyze human behavior. Automation of human behavior detection can speed up the annotation and analysis process significantly. Using the Restaurant of the Future as a benchmark, requirements are formulated for the system. For any kind of automatic analysis, classification and annotation of the behavior of people from a video, a robust method is needed to detect humans in the scene. For an indoor environment, background subtraction is a well-known detection method, but it faces many challenges such as illumination changes in the scene and shadows casted by moving objects.

Prerequisites of the system
The Restaurant of the Future is taken as a benchmark to list the requirements on the system (hardware, software, environment and objects). For example, the system must be able to deal with varying illumination, both in intensity and color because of the presence of large windows (influence of sunlight), the large amount of light sources and the different colors of these light sources. Another requirement is that persons have no any additional sensors or markers to ensure a natural recording. To detect people, background subtraction uses a model to represent the background scene.
Deviations of the current frame with respect to this background model identify the foreground objects. Several background subtraction algorithms have been analyzed on its performance and accuracy. To handle shadows casted by foreground objects and illumination changes in the scene, a multiple view background subtraction method has been investigated, which uses geometrical information about the scene to detect foreground objects. Often background subtraction gives unreliable or incomplete silhouettes. An alternative method based on level-sets has been investigated to find the contour of the object. The first results show that this method is less sensitive to shadows and illumination changes.

From detection to pose and gesture recognition
Ones the detection of persons is done, the persons have to be tracked and identified over time. The main challenge of tracking is (self)-occlusion, which can be handled by using multiple cameras. When the humans are separated, pose estimation can be performed on each person individually. This will result in the position of each body part, like the head, hands, shoulders, legs and torso. The final part of the analysis toolbox will be the recognition of certain pre-defined gestures. The software has to be tested and verified. The easy way is to use publicly available data sets for tracking and pose estimation. However, to have more freedom in our testing (specifically in gesture recognition), a data set will be recorded in the Restaurant of the Future and a ground truth will be provided.

Partners
Utrecht University
Noldus Information Technology BV

Contact details
Remco Veltkamp, Utrecht University
R.C.Veltkamp(at)uu.nl