Face recognition is one of the visual tasks which humans can do almost effortlessly while for computers it is a difficult and challenging task [1]. The applications of face recognition are increasing in a number of domains. An
upcoming application domain is user identification as a form of ambient intelligence for access control as an alternative
for pincodes and for adapting parameters of machines such as PC settings. Recently, a rapidly growing demand is for face recognition as part of a surveillance system. Nowadays most face recognition systems don't work at video speed but use previously captured video. Some huge systems are able to do on-the-y face recognition (matching detected faces to a limited database of stored faces) from captured video streams [2]. These latter systems are in high demand especially seen the latest
mondial political situation. Because of the cost of these
systems they are only affordable for large sites such as
football stadiums or airports.
What we want to show in this publication is that it is
possible using thought-over smart camera architectures to
achieve good, real-time face recognition results. A smart
camera is hereby dened as a stand-alone programmable
device with a size equal to or smaller than a typical video
surveillance camera. In our situation it is programmed in
such a way that video goes in and names of recognized
people come out.
Face recognition is becoming an important application for
smart cameras. However, up till now, the processing required
for real-time detection, prohibits integration of the whole application into a small sized, consumer type of
camera. This paper showed that by:
1. Proper selection of algorithms, both for face detection
and recognition,
2. Adequate choice of processing architecture, supporting
both SIMD and ILP types of parallelism,
3. Tuning the mapping of algorithms to the selected
architecture,this integration can be achieved. We implemented the algorithms
on a small smart camera. As a result we can
recognize one face per NGJms, when we are searching for
persons, with LIE% recognition rate and only Q% failure
rate.
Future research will focus on further tuning the mapping
of the algorithms, e.g. by replacing oating point operations
with xed point, trying other (cheaper) activation
functions (see eq. 8), and further parallelization of the
RBF neural network. This should allow for further speedups
needed when searching in much larger databases that can
contain large numbers of identiable faces. Furthermore,
the recognition will be enhanced by using multiple cameras
with different viewpoints.