|
Here's
a general overview of the whole system. The flowchart
shows how the process is going to be conducted from
the input from the camera to whether access is granted
or not.
First
of all, since the camera is facing a static background,
is it easy to store a mean representation of the empty
background. Therefore, the preprocessing part of the
system will take incoming frames from the camera,
and do a simple background substraction to retrieve
the foreground (human figure). Then, a simple heuristic
was developed to quickly detect the position of the
head, and then the position of the eyes is found to
extract that portion of the image.
The
face image (only one needed) and the eye portions
(multiple ones on multiple frames) are sent to two
different part of the system. The face image was sent
to the Neural Network to compute its features and
compared it with our face database. In the mean time,
the eye portion for every incoming frame, is filtered
and converted into a binary representation for future
encoding.
Finally,
for access to be granted at the end of the system,
the face has to be detected in the database and the
generated binary code has to match the corresponding
code for the detected person.
For more detailed information, please visit the site
I made for this. Please go check out the documentation
part of it, for the write ups and everything.
-[click
here]-
|
