The truths

Public misconceptions about AFR remains high, but the facts speak for themselves.

800 positive matches reached

AFR ‘Identify’ allows officers to load images of persons of interest and compare them against our 500,000 custody images to see if there is a possible match on identification.

AFR ‘Locate’ is a real time solution which utilises live feeds from CCTV typed cameras based either at specific, fixed locations or cameras secured to the top of one of our vehicle fleet to locate persons on prescribed watch lists.

76% accuracy at Wales vs Italy

During the Wales Vs Italy Six Nations rugby test match in 2018 the University Police Science Institute based at Cardiff University and South Wales Police tested the accuracy of the AFR Locate deployment in Cardiff City Centre. The trial involved 7 volunteers walking past the AFR cameras on multiple occasions with various walking styles and head garments.

The system detected these individuals in excess of 76 percent of occasions that they walked past the camera. This was for a false alert rate of 1 in 2,251 faces seen (see more on how to measure the accuracy of an AFR Locate deployment below.)

As you would expect props worn by the volunteers in particular hats and glasses that obscured the face impacted on the system being able to detect the individual.

In summary the system had an extremely impressive accuracy rate combined with a low incorrect alert rate.

Learn More

Who we are looking for

The system works by means of a pre-populated Watchlist which will contain information and images of subjects and a pre-defined response should these subjects be located by the system.

Watchlists will be both proportionate and necessary for each deployment with the rationale for inclusion detailed pre-event in the AFR Locate deployment report.

Primary factors for consideration for inclusion within a watchlist will be watchlist size, image quality, image provenance and rationale for inclusion.

A Deployment Report must be completed prior and finalised post deployment, this document will summarise the intelligence profile for an event, detail the rationale for deployment as well as rationale for watchlist composition. This document should also be used to capture lessons learnt from each deployment.

How accurate is AFR?

To date no individual has been incorrectly arrested as a result of an AFR deployment with no formal complaint made against its use.

Configurable alert rates

An alert will be received based on a threshold setting (sensitivity setting) of the system, i.e. how similar is this person to the person you are looking for - configurable by watchlist and by individual.



The lower the setting - the more incorrect alerts

The higher the settings - the less incorrect alerts

low setting graphic
high settings graphics
zero false arrests
0 False Arrests

The accuracy of a AFR Locate deployment can be measured by how many individuals located in your watchlist that are seen by the system in turn do not lead to an alert (i.e. walk past the camera without the operators being alerted) for any false positive rate.

Please see resources section for the Layman's Guide to the accuracy of an AFR Locate deployment.