On Aug 16, 2013, at 9:52 AM, Ian Walls wrote:

> Suma is the most practical and reliable way to do this right now, I think.
> 
> I've been investigating using a sensor network, but there are a lot of
> limits on the accuracy of PIR, and trip-lasers are low enough and require
> enough power that they'd be troublesome to maintain in a busy undergraduate
> environment.
> 
> One idea was to use an array of sensors:  PIR for motion, microphone for
> noise level and piezo/something similar for vibration.  The thought is that
> elevated levels of these 3 measurements should correspond to "high
> activity".  The placement and calibration of the sensors, though, would be
> key, and you'd need to do some thorough spot checking with Suma or something
> similar in order to be confident that what you're measuring (motion, noise
> and vibration) actually correlate to number of people.
> 
> The sensors would also need to be made out of cheap enough materials and use
> low-congestion wireless frequencies in order to be practical.  Balancing
> this with accuracy may never happen... but it would certainly be a fun
> experiment!


If you're going to take the sensor approach, and it's just a matter of
if there are bodies in specific places, you *might* be able to do it by
modifying cheap webcams.

Many are sensitive in infrared, so you take the IR filter out, and then
add a visible filter.

Position the cameras so that you have coverage of the area you care about,
have them take a picture at whatever times you care about, and then it's
just looking for hot spots.

(although of course, if you do this, it'd be just as easy for someone
to review security camera footage, if you have coverage in the places
you care about; the IR might be easier to automate the counting, though,
if you have someone who's good with automated image analysis)

And if it's just a matter of activity counting -- you might be able
to see if your wireless access points can tell how many items they're
in contact with, and use that as a proxy.

-Joe