Intelligent Real-Time Closed Circuit TV-based Surveillance, Automated
Verification and Access Control
by Shaogang Gong
A project on Intelligent Real-Time
Closed Circuit TV (CCTV)-based Surveillance, Automated Verification and
Access Controlhas been funded by the UK EPSRC Integrated Machine Vision
initiative and carried out at Queen Mary and Westfield College (QMW), London
since 1995 and due to complete in 1998. The project has developed an integrated
real-time system for automatic detection, zooming and following, acquisition
and recognition of moving faces in natural scenes.
The work has been motivated by a huge industrial demand in recent years
for CCTV camera-based systems that can perform non-intrusive but robust
automatic visual tracking, verification and access authorization in natural
but known environments. The system been developed is to be commercially
exploited by a new venture to be funded by venture capital investments.
At present, whilst visual surveillance is mainly relied on the detection
and discrimination of alarm events based on simple video motion detectors,
visual verification and access authorization are largely based on biometric
techniques which involve hand, fingerprint, iris and voice recognition.
These systems are either unreliable or aesthetically unacceptable and inconvenient,
giving slow response and extremely constrained in their operations. This
project has been focused on developing automated visual identification
of face image sequences automatically detected and tracked by a closed-loop
mechanism using conventional on-line CCTV cameras. The system performs
face detection and tracking robustly in real-time under a wide range of
lighting and scale variations by combining hybrid visual cues including
colour, motion and facial appearance models.
Figure 1: Real-time face detection and tracking under large scale and
lighting changes, and of occluded or partially occluded moving subjects.
For verification and access control, the project has developed a system
that performs real-time face recognition under large changes in scale and
pose based on face image sequences instead of static snapshots. One unique
characteristic of the system is that the use of sequences as opposed to
snapshots not only eases the burden of recognition accuracy placed upon
the individual face image frames (fault tolerance), but also makes use
of temporal continuity as a powerful additional computational constraint
for speed and consistency (temporal prediction).
The research carried out in this project has led to a few commercial
developments. These include a real-time intelligent people watching system
for surveillance in restricted areas, a one-to-one verification system
for ID and access control, and a group identity verification system for
criminal intelligence and crime prevention.
Figure 2: Real-time face verification and recognition based on captured
sequences of face movements using on-line adaptive learning and temporal
prediction.
This research has benefited from discussions between the QMW group and
Bond Security Services Ltd (UK) and Datacube UK Ltd (UK).
For more information, see web page at: http://www.dcs.qmw.ac.uk/research/vision/index.html
Please contact:
Shaogang Gong - Queen Mary and Westfield College
Tel: +44 171 975 5249
E-mail: sgg@dcs.qmw.ac.uk
