Automatic Robust Face Detection on High Resolution Smart Camera
Speaker: Yasir M Mustafah, ITEE
When: 2008-05-09 14:30:00
Venue: 78-420
Host: Prof Brian Lovell
Abstract:Modern society relies on video surveillance systems in ensuring safety and
security. However, current video surveillance systems are being used only as
forensic tools and not necessarily as automated real-time alerting systems.
Smart camera, a newly emerging technology, would be useful in order to
improve the current surveillance systems and to make intelligence
surveillance a reality. Recognizing faces in the crowd in real-time is one
of the key features that will significantly enhance Intelligent Surveillance
Systems. Using a high resolution smart camera as a tool to extract faces
that is suitable for face recognition would greatly reduce the computational
power requirement of the main processing unit. The processing unit for the
face recognition would not be overloaded with high resolution frames and
could be designed solely for face recognition.
Implementing a robust face detection system on an embedded smart camera
platform is a very interesting challenge. While there are some works that
has been done to implement face detection on embedded systems, so far there
is no implementation that is robust and fast enough to cater the demand of
surveillance systems that requires real-time response in high resolution
video.
Face detection itself is a challenging research topic. A lot of research
have been done for over a decade in finding the best technique to detect a
face in an image. One of the best techniques is the Adaboost based face
detection by Viola and Jones. Their method has attracted attention of many
researchers in face detection field. Recently, a research group in NICTA
introduced histogram feature as an alternative to the Haar-like feature used
in the Viola-Jones detection module. The histogram feature was designed to
be more efficient for embedded implementation.
This talk will report on preliminary work of developing of a prototype
smart camera that is suitable for face detection task. The camera utilises a
very high resolution image sensor suitable for crowd surveillance, and an
FPGA platform to allow rapid prototyping work. The talk will also reported
on the work done to improve the Adaboost based face detection module for a
better high resolution operation and to ease the embedded implementation.
One of the improvements is to combine the face detection module with a
background subtraction stage and skin colour filter stage to significantly
reduce the number of search windows in the high resolution image. The
software simulation result shows that this proposed method is much faster
compared to the original method while maintaining its accuracy and
robustness.
Biography:(biography unavailable)
Type: PhD Confirmation
Contact:Prof Brian Lovell, seminar host (lovell@itee.uq.edu.au)
or Guido Governatori (ITEE seminar co-ordinator)
(guido@itee.uq.edu.au)