Tip:
Highlight text to annotate it
X
Human computer interaction is quite an interesting topic for us.
It's been around for a while but I think it's only now that it's really starting to take
off and our interest is in developing more intelligent, autonomous devices that are able
to understand and interact with humans in a way that's useful for humans.
We initially started using photometric stereo for industrial applications where we were
looking to find defects on difficult surfaces such as this. This tile contains a deep three
dimensional topography which is camouflaged by this two dimensional coloured pattern.
And so what we are able to do is isolate the two and then quantify them.
After we did that we started to realise that maybe there could be other applications where
we are interested in topography as well, one of which is in skin cancer. So we developed
this device which also uses photometric stereo and we've been using this for a number of
years with the pigmented lesion clinic. The idea here is to look for a correlation between
disruptions in the skin-line pattern -- the pattern of polygons that we all have on our
skin -- and the presence of melanoma. Okay this is the Photoface device -- our 3D
face capture and recognition system. As somebody walks past this ultrasound trigger it fires
off these five near-infrared light sources and a synchronous camera acquisition. From
this we can estimate the surface orientation of each pixel and from that we can integrate
to form a surface. And what we want to do here is to build a
demonstrator system that can recognise demographic data so for example it can tell whether a
person is male or female, their gender, roughly their age and maybe even how interested they
are in advertisements that are being displayed on an electronic board. And then maybe alter
those advertisements to make them more interesting and more useful for people so we don't get
bombarded with irrelevant material. Okay so what we are doing here is applying
the photometric stereo method to 3D facial expression recognition. What we are specifically
looking for in this work is subtle expressions which occur naturally. So the experiment has
two components -- the first part of the experiment is to capture natural subtle expressions that
occur in everyday situations and the second component is then to analyse the prototypic
expressions which tend to be exaggerated and we are developing algorithms to recognise
both subtle and posed expressions. So what we do here is capture the evolution
of changes in facial shape. And we've been working with a group at Bristol University
to see if we can use this to detect depression in some of their patients. So the idea here
is that the way a depressed person may smile will be different from somebody who isn't
depressed, so the evolution of their smile, the time that the smile develops and the way
it develops across the face.