Subscriber Discussion

For A Potential Customer - Video Facial Recognition To Open/Unlock Door

MT
Matt Transue
Oct 12, 2016

We have a potential customer with an interesting request.

They would like to leverage video (assume new cameras) at the entrances of a residential building.  They would like facial recognition to identify and approve/deny access to the building.

They have existing access control but we are unsure who's, how old, etc.  We are still in the information gathering stage.

I know there are a TON of variables that make this difficult up front, but I am curious what you folks have seen or used in the past that may provide a potential solution for them.

We are in process of vetting what they really need vs. what the want. 

Any insight into facial recognition working with access control would be appreciated.  Thank you!

Avatar
Greg Thornbury
Oct 12, 2016
Facility Solutions Group, Inc. • IPVMU Certified

Take a hard look at Stone Lock Global. It relies on near infrared so lighting is not an issue like it is on some other facial recognition products that we've used.

It integrates with several access control platforms, and can also operate stand-alone at the door. The faceplate design incorporates a reader/keypad so you can set the validation method.

It's a very impressive product.

(2)
MT
Matt Transue
Oct 13, 2016

Thanks Greg.

I will check them out. I really appreciate the recommendation.

UI
Undisclosed Integrator #1
Oct 12, 2016

Why facial recognition versus using retina or IRIS? You can use retina scanners at some distance that will give the same effect, except the products already exist.

MT
Matt Transue
Oct 13, 2016

We will cover the retina scan option with them. They specifically said 'video' facial recognition...but again, I think they're just looking for a 'cool' factor and don't understand what other options are out there. Thanks!

MM
Michael Miller
Oct 12, 2016

I do not have any experience with them but have you looked at FST Biometrics?

(1)
MT
Matt Transue
Oct 13, 2016

Thanks Michael,

I have not looked at them. But I will. Always appreciated!

MM
Michael Miller
Oct 13, 2016

Here is a clip I found with looks like actual footage from the system.

(1)
JH
John Honovich
Oct 13, 2016
IPVM

We did a post about an FST deployment a while back: Debunking Start-Up's FST21 Case Study

(1)
MT
Matt Transue
Nov 07, 2016

Quick update here.

We have looked at FST and Stone Lock Global. Right now, it looks like Stone Lock has the most appropriate solution for this customer. Their NIR solution I feel would work better in this environment than a photo based solution. Lack of appropriate lighting is a contributor to this decision.

However, we are still engaged with both companies.

Once we have made a final decision I will try to share some details, as well as some photo's/examples of the solution in action if the customer signs off on it.

Thanks to all for the great suggestions here. Always appreciated!

(4)
JH
John Honovich
Nov 07, 2016
IPVM

Matt, thanks for the update!

Does FST not offer an NIR option? And how was pricing? Were they close in price?

MT
Matt Transue
Nov 07, 2016

FST has not introduced an NIR option to us, but, I did not specifically ask. Maybe I will do that.

We haven't gotten to pricing yet, but I suspect that will happen soon. I am still going to have FST quote their solution as well as Stone.

One I know more about pricing I will share. If I find anything out about an NIR option for FST I will update the thread as well.

(2)
UI
Undisclosed Integrator #2
Nov 07, 2016

I just got a demo from these guys about 2 weeks ago and the system seems interesting, but I am not sure I can think of a use case for it where it would be more reliable than standard access control. It is also fairly expensive, but I suppose you might be able to offset that price by using the same camera for alarm call ups, but it does not seem to integrate with any current VMS or IDS systems.

Good luck to them, but I am not a customer.

MT
Matt Transue
Nov 07, 2016

UI2, Stone integrates well with C-Cure 9000 (which is what this customer has). I am unsure about VMS integration.

FST is working on improved integration with C-Cure 9000 also, but it is not 'ready' yet.

The goal for this particular customer is not improved reliability of their existing ACS, but to add a 'seamless' entry aspect for their residents.

UI
Undisclosed Integrator #2
Nov 07, 2016

Yeah, we are a Software House facility which is why we were looking at it but even so... I have my doubts on how tight that integration will be.

MT
Matt Transue
Dec 07, 2016

Update:

We did an on-site demo with the Stone Lock Pro device.

It was unable to read a face more than 3' away from it, even in optimal lighting conditions (inside the garage with static lighting)

When we took it outside it did not read at all, even when about 1 foot away from the unit. This seems to be due to the sunlight that was being cast on the face of the individual as they were being read.

In fairness, they did warn against an outdoor solution with sunlight directed at the reader for the face of the individual.

The Stone Lock Pro will not work for this customer's requirements.

We also had FST out last week for a site survey and demo of their solution.

Their solution should work fine with the outdoor lighting they have. The concern is at night and in general low light conditions. They require a minimum of 200lux on the target.

We will likely need to add some lighting to the two entrance areas.

We are planning a presentation to the customer next week for them to decide on this solution.

Thanks again for all of the helpful suggestions here. I will circle back if/when the customer accepts the offer and we move forward with the solution.

(1)
JH
John Honovich
Dec 07, 2016
IPVM

Matt, thanks, very informative!

So did FST's face recognition work better or?

MT
Matt Transue
Dec 07, 2016

It seems that indoor, low light conditions are where Stone Lock Pro excels. They use NearIR technology so they require no light to work, however, they perform poorly in brightly lit areas (such as many outdoor entrances). The detection range is also lower or closer than FST. A subject can be no more than 3' away (from our tests...Stone will say 4' max, but that didn't work reliably for us). Admittedly, Stone Lock does not have a true outdoor rated solution. We would have had to 'protect' the reader from the elements in some way if we wanted to use it outside.

Because of these issues Stone was not a good solution here. In low lit areas where the subjects will be close to the reader, I think their solution would be very good.

Subject detection on the Stone Lock Pro was about 1 second.

For FST, they perform better in areas that are well lit, but struggle in low lit areas. They require a minimum of 200lux for a proper facial read. One advantage of the FST solution is the subject distance is longer than Stone Lock.

We had one of their engineers out to the site last week and based on what he saw, he had no concerns on the outdoor entrance. However, we do need to add light to both the front outdoor entrance and the garage entrance so that at night there is enough light for a proper read.

We should be able to get a good read about 6-7' from the camera which is much more realistic for the front entrance.

We were able to take some shots in the vestibule, but not directly outside...but the engineer was not concerned with it, so I'll take them on their word for now.

To answer your question directly, if conditions were such that the Stone Lock Pro would work correctly, I would likely choose that solution over FST. I personally like the NearIR (non-photo based) solution better. Detection was a bit quicker (Stone = about 1 second, FST = about 2 seconds).

However, that is not to say the FST's solution is not good. It seems to be very good, just different.

I don't think I'll know how well it does or does not work until we actually deploy it and begin registering residents and seeing how it performs in real life.

MT
Matt Transue
Dec 07, 2016

I don't have a picture of the FST setup yet, but I do of the Stone Lock Pro device.

(1)
U
Undisclosed #3
Dec 07, 2016

Open in Paint - Rotate right 90 degrees - save as.

Just sayin'.. :)

Avatar
Brian Rhodes
Dec 07, 2016
IPVMU Certified

That appears to be a page issue, not operator error. I'll file it as a bug.

(1)
MT
Matt Transue
Dec 07, 2016

UD3 - Fair enough :)

But it's the correct orientation on my PC (5312 x 2988), so no rotation necessary natively. If I rotate it from the native image I will rotate it to the wrong orientation.

MT
Matt Transue
Dec 07, 2016

That's 5312 High and 2988 Wide

U
Undisclosed #3
Dec 07, 2016

yeah just messin' with you - I had to shrink it down a bit first, then rotate.

I'm a Paintophile :)

MT
Matt Transue
Dec 07, 2016

No sweat!

I'm a big fan of calling a spade a spade (or a dumbass a dumbass in this case) :)

U
Undisclosed #3
Dec 07, 2016

(1)
MT
Matt Transue
Dec 07, 2016

Nice

MT
Matt Transue
Dec 21, 2016

Update:

We have submitted our proposal and presented to the executives.

FST is the solution we went with, utilizing 3 Axis cameras.

  • Q3505 22mm
  • Q3505 9mm
  • P3367 9mm

The vestibule camera will be approx. 11 feet away from the approach path.  FST requires a minimum of 100ppi between the eyes, thus requiring a pretty hefty lens (22mm).  The FST engineers have confirmed that the 22mm option on this camera will provide proper recognition 11 feet away.

We will need to improve the lighting in the garage.  We will also need to improve the lighting for the two main entrance cameras if they want to use them at night.

Thank you all for your feedback and assistance here.  It is greatly appreciated.

(1)
JH
John Honovich
Dec 21, 2016
IPVM

Matt, thanks!

One question. You say "FST requires a minimum of 100ppi between the eyes" Can you clarify? That's a 100 pixel per inch or? Or do you mean 100 pixels total between the eyes?

MT
Matt Transue
Dec 21, 2016

Sorry John, I was pretty vague and misleading.

100 pixels-per-inch minimum (doesn't relate to a particular 'position' on the face)

They initially look for data 'between the eyes', that's how it was explained to me by the FST Engineer.

They also have a learning algorithm(s) so over time the system will learn 'behavior' of a subject.

So within the read time (less than 2 seconds once in the approach path) they must reference the data between the eyes as well as what has been 'learned' about that subject.

I am speculating a bit for now.  I hope to know quite a bit more when I actually get to set up the system from scratch.  Unfortunately, I was unable to get a demo kit in house yet so I haven't played with it myself...but I have seen the demo.

JH
John Honovich
Dec 21, 2016
IPVM

I think that's 100 pixels between the eyes. If it was 100 pixels per inch, that would be 1200 ppf which on a 1080p camera would equate to a 1.6 foot FoV (1920 horizontal pixels / 1200 ppf). Also 'pixels between the eyes' is a fairly common metric for facial surveillance. Some go as low as ~40 pixels between the eyes but systems that want higher performance tend to go in the ~100 pixels between the eyes range.

MT
Matt Transue
Dec 21, 2016

I will double check, but I think it's 100 pixels per inch (or, 1.6' HFoV).  I recall asking them "What is the minimum pixel per foot requirement for a valid read" and they answered, "100 pixels per inch".

All of the demo stuff I saw utilized a very tight FoV also, likely no more than 1.5' HFoV.

They do tend to push the 5MP cameras, maybe this has something to do with it.

Let me verify with them just to be certain.

UM
Undisclosed Manufacturer #5
Nov 13, 2017

For 2D, NIST calls for minimum of 120 pixels per inch for intra-eye distance.  Might be part of in ISO/IEC 19794-5.

From a open session at ISE West this year, thought I heard that minimum of 50 dots/per inch is good enough for 3D facial recognition.

MT
Matt Transue
Feb 08, 2017

All, I want to thank everyone again for the helpful comments in this discussion.

Unfortunately, the customer has decided not to go with the proposed solution at the moment, so this is stalled/lost.

They did like the 'cool' factor of the solution, but don't have it in the budget.

@John, I did follow up with the FST engineer regarding 100ppi or 100 pixels between the eyes.  It is, in fact, 100 pixels between the eyes as you eluded to earlier.

Again, thanks to everyone for the comments and suggestions. I wish I could have shared a success story on this one.

Cheers!

(1)
UI
Undisclosed Integrator #4
Nov 13, 2017

Thanks Matt for your updates.

I have a client that is looking into using this for access control.

I'm currently looking at Allevate's product and will commence testing shortly to see how effective it is.

The only downside is that it is pricey, for what is effectively "cool" access control.

Jason

MN
Mike Newman
Nov 16, 2017
IPVMU Certified

Thanks Matt, for keeping us updated on your process with FST.  We recently engaged them as a possible solution.  Has anyone deployed FST that can provide me some feedback on their performance?

UI4, did you look at FST along with Allevate? 

UI
Undisclosed Integrator #4
Apr 23, 2018

Hi Mike,

We haven't tested FST just yet but are about to.

The Allevate solution worked well but we believe we need an "on-prem" solution due to the issues around latency.

Issue wouldn't be as relevant if it was just for CCTV review, but as we plan to use it for access control, purchasing an MX server may be cost prohibitive.

We'll test the FST solution next and take it from there.

Thanks,
Jason

SD
Shannon Davis
Apr 23, 2018
IPVMU Certified

StoneLock is the best innovation, and it actually works as described, device I have seen in my over 25 years in this industry. For bio-metrics, IMO, there is nothing better. They are still a little pricey to some but once a customer has one in operation there is nothing better in terms of the security it provides and functionality. Also the good news is they have started the process to begin manufacturing in the United States to meet the buy American Act.

New discussion

Ask questions and get answers to your physical security questions from IPVM team members and fellow subscribers.

Newest discussions