23-07-2012, 11:57 AM
Extract feature points from faces to track eye's movement
Extract feature points.pdf (Size: 1.54 MB / Downloads: 102)
Abstract
This paper handles with the question how to extract xed feature
points from a given face in a real time environment. It is based on the
idea, that a face is given by Viola Jones Algorithm for face detection and
processed to track pupil movement in relation to the face without using
infrared light.
Introduction and Idea
There are a lot of projects running all over the world trying to track faces and
points in it. Our target is to track eye movement in correlation to other xed
points in the face without using infrared light. Infrared tracking of pupils is the
current state of the art, but it has some disadvantages as it can cause damage
on eyes.[1] Further it is necessary to have infrared light, which can be reected
in the eyes.
In contrast to infrared light tracking, it is tried to follow eyes movement by
extracting dierent features from a standard webcam picture and process the
given data in a real time environment.
Environment
It is necessary to dene two dierent environments. First of all the real world,
secondly the programming environment. The rst one denes where the webcam
actually is, the second one, where the system is running on.
Real world environment
There are a lot of inuences, which can manipulate the result of tests in a
signicant way. All tests are made under normal indoor light conditions,
which mean that there is no extreme light coming from sides. Of course, there
must also be enough light in the room, otherwise a visual tracking can not work.
Best results are made with light from above.
Programming environment (Soft- & Hardware)
For tests and rst implementations Matlab R2009a[2] was used. It is not as fast
as C code id, but not fast enough for real-time experiences. Another advantage
of Matlab onWindows machines is the possibility to simple access webcams with
the Image Acquisition Toolbox[3].1 On the hardware side a Logitech Quickcam
Pro 9000[4] is used for image acquisition and a Lenovo ThinkCentre 639471G[5]
(Intel® Core 2 Duo E7200 Processor, 2 GB RAM, Windows Vista) for testing.
Nose holes
Finding nose holes in an area given from face's geometry depends on the angle
between camera and face. If there isn't a direct line of sight between nose holes
and camera, it is obviously not possible to track them.
Nose holes' color have a signicant saturation, depending on its color black.
The threshold must be dened and over geometry or clustering two centers of
saturation can be found.
Mouth
Detecting the middle of the mouth isn't as simple as it is thought. There are
a lot of possibilities, going over gradient horizontal and/or vertical decent, hue
or saturation. At the moment it is implemented utilizing the distinct hue of
lips. Light reects on lips and this point is fetched by a dened hue value. In
contrast to the methods, this method is not light independent, thus intensity
and direction of the light can inuence results. A better method should be
included in the future.