Subscriber Discussion

NEC Facial Recognition UK Festival Results

U
Undisclosed #1
May 09, 2016
IPVMU Certified

Came across this FOIA request regarding a festival that was held last year in England.

Apparently the idea was to place cameras around the festival area, and to capture images of the 100,000 festival goers and the. match against a subset of known festival criminals / cell phone theives.

No matches, or at least no arrests or evicitions resulted from the system, although the control cops were picked up 77 times.

People weren't told ahead of time that their picture was to be used in an automated face rec system.

Some gratuitous detail supplied by the authorities:

This works by scanning faces passing CCTV cameras and comparing each one against a watchlist. Each facial image is immediately compared against the watchlist and if not matched, the image is removed from the system. However, if a person of interest is identified, NeoFace Watch immediately sends alerts to officers and staff. The potentially matched images were only retained for the duration of the festival.

Intelligence officers drew together a database of offenders from across Europe who were known by the police to target large music festivals to commit crime. Their images were enrolled into the NeoFace Watch system, provided by NEC. This was a standalone system located within the festival grounds and not connected in any way to any external system.

Six cameras were placed in a small area of the site at Donington Park, for some of the festival, and scanned the faces of festival goers. At no point during Download were unmatched facial images retained by the system. Because it was a controlled trial, a number of officers and staff volunteered to have their photograph enrolled into the database and they were successfully picked out 77 times.

No one was arrested or evicted from the site as a result of the NeoFace software. The database of offenders and potential matches was permanently deleted at the end of the event.

Related: 18 Million Face Database Now Operating In The U.K.

JH
John Honovich
May 09, 2016
IPVM

1, thanks for sharing, very interesting.

No matches, or at least no arrests or evicitions resulted from the system, although the control cops were picked up 77 times.

The control cops being up 77 times is no big deal, one way or another. (1) They were enrolled right then and there, which means similar look / time and probably a high quality camera and direct pose. (2) Those cops certainly walked past the six cameras numerous times so you would expect them to be picked up a lot.

The other point is harder to interpret. Does that mean that there were no matches against that "database of offenders from across Europe who were known by the police to target large music festivals to commit crime" or does it mean that they got correct hits but simply choose not to arrest or evict those known criminals?

One interesting point that would be great to know is how many times did the 100,000 festival goers incorrectly match the hit list of known criminals? This is generally a big issue. Did they get 10 false matches, 100, 1,000, 10,000? It would be great to know this to better assess the performance.

I will give it to NEC, they are obviously spending a lot on this (e.g., this free system / trial) but I still wonder how much success they are getting.

(1)
Avatar
Brian Rhodes
May 09, 2016
IPVMU Certified

The vagueness of the response is a win for FUD (ie: Fear, Uncertainty, Doubt) playing a role in the usefulness of this deployment.

The actual performance of the system is secondary to the fact now both the Donington Download Festival and Leicestershire Police have established they use the system and offered up some ambiguous claims of effectiveness, ignoring hard success metrics.

The response makes it clear two different NEC products were used:

  • NeoFace Watch: Real-time analysis used at the festival
  • NeoFace Reveal: Forensic only, adopted by Leicestershire Police

I gather from this that real-time face detection of the city/area was never a goal, but the music festival presented a test opportunity. The risk for NEC and the city were small, because rumors about the deployment of a face-rec system in the area would be win enough.

However, the win: At this point, will not criminals seek other venues and locations to commit premeditated crimes?

KA
Konstantin Avramenko
May 09, 2016

From my experience, the first thing such kind of vendors ask you to do is to sign CDA or similar document where all test results are mentioned as the confidential information and it is strictly prohibited to share any data without the written consent from the vendor side. I could test them but all I could say officially to customers was that in this particular case vendor x is better than vendor y without any numbers. Thus it is difficult to expect to get any real results from such kind of tests.

U
Undisclosed #1
May 09, 2016
IPVMU Certified

That's why it's odd that they offered the additional detail about 0 arrests and 0 evictions when they weren't even asked.

KA
Konstantin Avramenko
May 09, 2016

No one was arrested or evicted from the site as a result of the NeoFace software. The database of offenders and potential matches was permanently deleted at the end of the event.

It sounds like a disclaimer that it was just a test and all data was deleted after the test.

JH
John Honovich
May 09, 2016
IPVM

the first thing such kind of vendors ask you to do is to sign CDA or similar document where all test results are mentioned as the confidential information

They can certainly ask what they want. On the other hand, this is being done with the government and the government has certain obligations / responsibilities about disclosure of government actions.

New discussion

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

Newest discussions