Re: Re: Room occupancy detection

["Peter and Jane" <pandj.munnoch@xxxxxxxxxxxxxx>]

How about thermal imaging? Unless the wearer dies(!) they will
register warm while the coat on its own won't. Heat sources like radiators
could be learnt.
Just a thought.
Rgds
Peter
----- Original Message -----
From: Mal Lansell
To: ukha_d@xxxxxxx
Sent: Wednesday, December 03, 2008 12:37 AM
Subject: [ukha_d] Re: Room occupancy detection


--- In ukha_d@xxxxxxx, Kevin Hawkins <yahoogroupskh@...> wrote:
>

> Overall though I'm not sure that BT is the right way - as there are
> always phones left on a table or people without phones, plus I rarely
> carry mine. I still feel a carefully written camera image
analysis
> (IR sensitive ?) based application that handles appropriate sized
shapes
> that appear in the field of view coupled with movement is a good
basis ,
> coping hopefully even with sleeping individuals on a sofa. Yes it
might
> get thrown by the occasional discarded coat (but so could BT).
> Changing light conditions and shadows I'm sure are going to be hard
to
> handle.

I think your right about cameras being the best way forward, although
it takes quite a bit of processing to analyse a frame. A neural net
can be trained to differentiate between occupied and not, but again you
get the issue of a discarded coat and you never really know what you
have trained the net to recognise (a classic example is the London
Underground - the net was trained to recognise a full platform, but it
kept giving false postives. it turned out that there was a bright
shiny bin in the corner, which when hidden triggered the "platform
full" response, whether there was one person blocking it from view or
a
whole crowd!)

I've done some research on this with the aim of detecting cats in the
garden and turning on the sprinklers to chase them off. It's fairly
easy to detect moving objects (pixels that remain the same over a long
period are classed as "background", everything else is a
potential
target). Movement from shadows and changes in sunlight are dealt with
fairly well by converting the image from RGB to HSL - the hue doesn't
change much with varying brightness.

Still a way to go though, but there is potential to use a similar
system for room occupancy (even more so actually, since a coat won't
move at all, whereas a person will move a bit, even if only slightly)

Mal





[Non-text portions of this message have been removed]


------------------------------------

********************************************
Limited Time Offer - Free iPhone VAT Calculator
http://berble.com/index.php/component/content/article/109
********************************************