infrared sensor array for people counting

[p_stoyanoff]

Hi! I need infrared sensor array suitable for people counting. I think 16x16 or 32x32 sensor matrix is enough. Does anyone know where I can buy such sensor from.
Thanks!

 

[CocaCola]

IR is not real reliable for counting people, unless they are in a constantly moving single file line, arms to their sides, no extreme leg movement and spaced out from one another to name a few things...

IR is better left to detection not counting in most applications, unless it's a controlled environment...

 

[p_stoyanoff]

Take a look at this video:

I need something similar. I tried to connect with this pyreos but they didn`t return a decent answer. Do you know another company who has similar products??
Thanks!

 

[CocaCola]

You can get detection like that with pretty much any basic PIR sensor, I would just use a micro to debounce the detections... That is if you don't need the fancy computer interface, that would be more work...

But, what I stated before still applies, notice they all walked at a near constant pace nice and spaced out? That is so it works properly... It's a very unreliable way to count people in reality...

 

[(*steve*)]

It might be even easier to use an optical mouse. The sensor in that has more resolution and is easily interfaced to a microcontroller.

More resolution beans you can have people bunched up a bit more, but yeah, as CocaCola says, people don't walk along in single file, nicely spaced out.

 

[p_stoyanoff]

Can you offer me such sensor. I have to monitor area 3x3 sq . m. from the top.
Thanks!

 

[CocaCola]

Can you offer me such sensor

Maybe, I have no idea of your design requirements... As I said any PIR sensor will do the detection, it's all a matter of how 'fancy' you want to get on that detection and what type of interface you need that will dictate specifics...

Something like this

http://www.ebay.com/itm/1-Pcs-Adjus...865?pt=LH_DefaultDomain_0&hash=item1e696389a9

will do the job, if you do as I suggest and use a micro to debounce the detections... But, as I keep saying only if the people play along... Two people holding hands, hugging or even walking side by side (close) will likely result in a single count no matter how many there are... On the flip side people waving their hands, moving at irregular/erratic paces, or moving back and forth in the target area will almost certainly result in multiple detections...

 

[p_stoyanoff]

Two people holding hands, hugging or even walking side by side (close) will likely result in a single count no matter how many there are... On the flip side people waving their hands, moving at irregular/erratic paces, or moving back and forth in the target area will almost certainly result in multiple detections...

This is what I want to avoid! I need a sensor constructed with more then one cells. For example 16x16, 32x32. This will allow me to "see" the area where something is moving and if it`s too large I could count it as 2 persons. Other advatage will be many people can pass simultaniously and I`ll be able to detect almost all of them. See what I meen now?

 

[CocaCola]

This is what I want to avoid!

You can't with PIR, you can try but you will fail... You simply don't have the resolution or resolve level to accurately avoid these issues...

I need a sensor constructed with more then one cells. For example 16x16, 32x32.

So grid multiple sensors...

This will allow me to "see" the area where something is moving and if it`s too large I could count it as 2 persons.

200lb adult
multiple 50lb children in close proximity
child glued to adults side

Those will look nearly identical to a PIR detector...

You also have to consider clothing, a person wearing a winter coat will leave less of an impression than someone in thin clothing... Ambient temps will also wreak havoc with impressions...

Other advatage will be many people can pass simultaniously and I`ll be able to detect almost all of them. See what I meen now?

I suspect once you actually implement it you will find out it's not as accurate as you envision... Yes, you can make it more accurate with algorithms trying to organize and analyze the detections, but regardless you simply don't have the resolution of data needed for 'accurate' measurements...

The video you linked to is a proof of concept that will work in a controlled environment, but in the real world the results will be quite different...

 

[p_stoyanoff]

This is not the point! I need to buy a suitable sensor. I need a company which produces such sensor or a store where i can find.
Can you help me with that?

 

[davenn]

This is not the point! I need to buy a suitable sensor. I need a company which produces such sensor or a store where i can find.
Can you help me with that?

it is exactly the point

Why is it that you refuse to listed to excellent advice ?

You are most likely not going to find a sensor to do specifically what you want to achieve unless you make changes to the way people will pass by the counter

Dave

 

[CocaCola]

As davenn stated already this is the point, you are asking me to help find you something that honestly doesn't exist at a workable level as you have specified for what you want to do... What you saw in the video is just a proof of concept, not a finished product that actually works and is reliable in the real world, regardless of their claims I wouldn't hold my breath...

If you want 'accurate' IR detection for things like counting or gestures you need to get away from a small resolution grid of detectors like that... Even at your desired 32 x 32 that is only 1024 pixels to work with and it's poor accuracy to begin with... You can get a 307,200 pixel (640 x 480) pin hole camera for chump change, that is 300 times the resolution... Or get into dirt cheap mega pixel cameras that are 1000s of times better resolution than your desired 32x32 grid... At that resolution you can get the accuracy you need, or at least reduce the error significantly...

Where do you get such a beast? Easy they are called CCD sensors and they can be found in digital cameras... Cheap modules like this...

http://www.ebay.com/itm/VGA-CIF-OV7...94?pt=US_Security_Cameras&hash=item27ca77c1e2

Or webcams like this... Even higher mega pixel resolution and no IR filter...

http://www.ebay.com/itm/New-USB-2-0...t=LH_DefaultDomain_0&var=&hash=item2a227d9f33

Depending on the lens you will want to remove the IR filter if it has one (many do because it distorts colors) some of the 'night vision' ones will not have the IR filter because they are using the IR... Next you will want to mount the camera behind a visible light filter (or swap the IR filter for a visible light filter)... Where do you get one of those you might ask? Easy go get a roll of new 35mm slide film, take it to a developer and get it developed unexposed, aka buy it and hand it right back in to get developed... Presto you now have a bunch of visible light filter material...

Next assemble this camera and interface it to your computer... If you skipped the camera module board and went right to a cheap webcam it's all setup, just plug and play complete with software to view the image...

You now have a 640 x 480 (or better) IR sensor array... Don't want that high of a resolution, simple grid it off into small blocks and average the image across each block to scale it down...