Of the many biometric approaches available for surveillance applications, face and gait biometrics are natural due to their noninvasiveness, i. e. acquisition of face image and gait sequence in principle requires no cooperation of the participants. Although face recognition has been actively studied over the past decade, the state-of-the-art face recognition systems yield satisfactory performance under controlled conditions and degrade significantly when confronted with illumination and pose variations, aging, expressions, disguises, etc. Gait-based human recognition, as an emerging biometric, generalizes inadequately across surface type, camera viewing angle, load carrying conditions and even shoe type!
The goal of this tutorial is to provide a comprehensive review of face and gait based human recognition algorithms with applications in surveillance. Specifically, we will discuss methods for face recognition using video sequences, illumination-invariant still face recognition methods, and face recognition across aging, gait-based human identification using fronto-parallel and arbitrary views.
The tutorial is accessible to a wide audience since only basic level of linear algebra, probability, statistics, and image processing is assumed. Graduate students and researchers new to the field can use the tutorial to quickly comprehend the state-of-the-art of unconstrained face and gait recognition. Also the tutorial could serve as a starting point for them to embark on their research on face and gait recognition. Designers of surveillance systems can readily extract useful engineering principles that will come in handy in their work.
S. Kevin Zhou, Integrated Data Systems Department, Siemens Corporate Research Inc., Princeton, NJ 08540
Rama Chellappa, Center for Automation Research, University of Maryland, College Park, MD 20742
- Introduction [pdf]
- Part I: Face-Based Human Recognition [pdf]
- State-of-the-art
- Face recognition under pose/illumination variations
- Face recognition from videos
- Part II: Gait-based Human Recognition [pdf]
- Gait-based human identification using appearance matching
- Statistical framework for gait-based human identification
- View-invariant gait recognition
- Combining multiple evidences for gait recognition
- Discussions