Subscriber Discussion

Best 1-N And 1-1 Facial Recognition Systems?

Avatar
David Carta
•Oct 25, 2018
Telaeris, Inc.

I'm looking for recommendations on both 1-N and 1-1 facial recognition systems for a client.

I'm interested in the following:

1. Pricing - often these systems are hugely expensive.
2. Licensing - perpetual if possible, yearly maintenance ok
3. False positive performance - matching live face to wrong person
4. False negative performance - not matching live face to correct person
5. Ease of deploy
6. Support by vendor

Clearly the 2018 NIST test is ongoing, but I'd love to hear some real world experiences.

For those interested, this is the 2017 IARPA Face Recognition

Thank you,

David

 

JH
John Honovich
•Oct 28, 2018
IPVM

David, thanks for the question. We have not done enough testing nor research to give a recommendation.

Since you are looking at the NIST reports, one that might be helpful for your 1:N request is the NIST FIVE test report. It's from last year and it covers video based non-cooperative scenarios applicable to video surveillance. We just recently started to review it so I can't net it out yet but even the first 20 pages are quite helpful in terms of giving key findings and guidance. If I am reading it correctly, generally NEC (coded M in the report) was rated most accurate in this test.

Roughly speaking you have 2 groups:

  • Premium priced non-Chinese: NEC, Herta, Anyvision, etc., who promise higher performance but at thousands per channel.
  • Chinese low cost: Dahua, Hikvision and most Chinese mid-sized video surveillance manufacturers who are either including FR for free or in relatively inexpensive cameras / recorders. From our testing of Dahua recently, it might work in a small office with optimal camera positioning but would be risky for bigger, more demanding applications.

Let me know if you have feedback or questions.

(1)
Avatar
Campbell Chang
•Oct 29, 2018
In reply to John Honovich

You forgot about a 3rd group, the web services? (seems like a good enough name for them) players.

Amazon Rekognition

Microsoft Azure

IBM Watson

Google Face

etc.

 

I have no idea how these perform or any of their features.

JH
John Honovich
•Oct 29, 2018
IPVM
In reply to Campbell Chang

Campbell, thanks. They are definitely worth mentioning. The practical issue is the latency. Sending this to the cloud for  1:1 (e.g., access control) or 1:N (e.g., real-time face alerts) would cause problems due to the relatively high latency (5 seconds or more to send the video or images to the cloud, have the web services analyze and then send back). 

If this was just being done for search / investigations, that would not be as much of an issue and could be significantly less expensive than setting up a whole system. Related: Amazon's "Dangerous New Face Recognition Technology" Says ACLU gives examples about face recognition search.

GG
Gautham Gopalakrishna
•Nov 05, 2018

@David - Did you get any information on the pricing? I'm interested to see if the NEC & Herta have started offering SaaS based solution or it's one time license + annual fees?

Avatar
David Carta
•Nov 05, 2018
Telaeris, Inc.

The information is still quite sparse and I'm digging into VisionLabs solution right now.  Sorry not to have more feedback for you.

New discussion

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

Newest discussions