Adaptive Systems for Face Recognition in Video Surveillance

Lecturer: 
Professor Eric Granger
Lecturer's institute: 
École de technologie supérieure, Université du Québec, Montreal, Canada
Date: 
2.9.2014 13:15
Place: 
TS127

 

 

Infotech Oulu Lecture Series

Adaptive Systems for Face Recognition in Video Surveillance

Lecturer: Professor Eric Granger, École de technologie supérieure, Université du Québec, Montreal, Canada

Date: September 2, 2014
Time: 13:15-15:00
Room: TS127

Abstract: Recognizing faces belonging to individuals of interest in video surveillance remains a challenging problem due to camera interoperability and to variations in face capture conditions over time. Prior to operations, the facial model of each individual is designed a priori using a limited number of reference samples.  Although these models maybe adapted when new reference images or videos become available, incremental learning with faces captured under changing capture conditions may lead to knowledge corruption. This presentation describes specialized adaptive multi-classifier systems for video face recognition (FR) in changing surveillance environments. To sustain a high level of performance, yet preserve previous knowledge, facial models stored in these systems are adapted according to the context, in response to new reference face images and trajectories. Experimental results obtained with the Faces in Action (FIA) and Chokepoint video datasets indicate that the proposed systems allow for scalable architectures that can sustain a significantly higher level of accuracy than reference FR systems, while minimizing computational complexity.
 

Biography: Eric Granger obtained a Ph.D. in Electrical Engineering from the École Polytechnique de Montréal in 2001, and from 1999 to 2001, he was a Defence Scientist at Defence R&D Canada in Ottawa. Until then, his work was focused primarily on neural network signal processing for fast classification of radar signals in Electronic Surveillance systems. From 2001 to 2003, he worked in R&D with Mitel Networks Inc. on algorithms and dedicated electronic circuits (ASIC/SoC) to implement cryptographic functions in IP-based communication platforms. In 2004, Dr. Eric Granger joined the ÉTS, Université du Québec, where he has been developing applied research activities in the areas of machine learning, patterns recognition, signal processing and microelectronics. He presently holds the rank of Full Professor, and is a member of the Laboratoire d'imagerie, de vision et d'intelligence artificielle (LIVIA). His main research interests are adaptive pattern recognition systems, incremental and on-line learning, change and context detection, computational intelligence, and multi-classifier systems, with applications in biometrics (face recognition and signatures verification),  Computer and Network Security (intrusion detection and watermarking of digitized documents), and surveillance (video analysis and recognition of radar-comm signals).

More information: Abdenour Hadid

Last updated: 15.8.2014