Subscriber Discussion

I Need To Count Prisoners, Help

A member has asked about counting prisoners. He's well aware of video analytic based people counting solutions. However, unlike a retail environment, where counting can be rough (i.e., today we had ~1500 people + or - 100) missing prisoners carries significant risks.

They have already considered RFID but that's dependent on the tag/chip staying embedded with the prisoner/clothing, somethig that the prison cannot depend on.

So, is there a prison specific automated counting solution or a very accurate, hard to defeat one that could be use?

Are installing turnstiles an option? It may not be useful for counting heads in a crowd, but full-height turnstiles installed at the entrance and exit pathways will establish exactly how many people have passed through into a room.

Maybe but I am assuming it can't be a physical barrier. How about an optical turnstile? Btw, here's your/our turnstile guide.

Another suggestion, albeit one I have no experience with: a thermal 'People Counter' sensor.

I noticed the spec sheet reads 'highly accurate!', but gives no real indication of how accurate (+/- n people) it typically performs.

Because we do "video analytics" I get asked this question *a lot*. Probably at least once a week at this point, in various iterations.

It's not something we do, and I can tell you a high accuracy solution is very very difficult, and I'm not personally convinced that ANYTHING exists that can do this in a manner that I would personally feel comfortable relying on.

As you can probably imagine, difficulty level goes up dramatically as density of persons increases, and then multiply that again if the persons are all dressed more or less alike.

If you can control the flow of prisoners, essentially controlling the spacing in between them somehow you *might* be able to get something workable. If you can control the spacing *and* they all dress in the same orange jumpsuit or whatever it could get even easier to count orange blobs. You're going to have to factor in lighting, local reflective surfaces and things like that.

Brian, what about thermal or 3d based sensors?

Haven't used or evaluated any of them first-hand to be able to say for sure. I've always personally felt that the ideal solution would be a camera/analytics + Kinect style setup. Those two devices, fed through the right processor should be able to yield very high accuracy and have the greatest chance of not mistaking 2 medium sized people for 1 large person, or 1 large person for 3 people, etc.

I would speculate that in an overheard view the thermal units could have issues with tightly grouped clusters of people. For a retail store it's probably not an issue. Very few customers are going to walk in right next to each other, and you have some natural spacing between the customers. A prison cattle-call line is going to be a much different problem.

I think there is one important variable in this scenario that has yet to be mentioned: the countees.

Once they figure out the 'counting devices' are there, prisoners will actively work to beat/defeat your efforts to 'count' them unless the counting process is highly controlled - as others have mentioned; spacing, single file lines, etc. If there are any ambiguities, expect these folks to exploit them - simply because they can.

Any solution must 1st technically 'work', but equally important in this application, also be highly 'game-resistant' since the countees in this environment having nothing but time to figure out ways to defeat your efforts. :)

I think Brian's and John's suggestion might have merit. I am researching the thermal solution now. Perhaps that plus the video analytic will be a strong start.

I would recommend you to consider the face recognition software called FaceVACS-VideoScan developed by Cognitec. Their newest version has some algorithms to count individual faces (among other features).

Wouldn't this be a good application of a fingerprint biometric scanner placed inside the room to be entered? That way you can tell not just how many can in, but who did not as well.

Bohan brings up an interesting option. I would hesitate to use a video-based solution in this application because of your need for high reliability. What about using turnstiles, opened with biometric readers (fingerprint or something like Techsphere's vascular scanners), that are associated with overwatching cameras?

wouldnt it be cheaper and more accurate just to hire someone(s) to count these people? throwing technology at it just may not be the answer?

Humans are subject to ADD/boredom and bribes.

Quote from Agent Smith: "never trust a human to do a machine's job"