The UK National Crime Agency (NCA) has published details of its plans to deploy facial recognition capabilities to devices without an internet connection.
It has released a request for information (RFI) inviting the industry to provide information about the delivery of advanced solutions capable of identifying individuals, whether they are alone or standing in a crowd.
The NCA is responsible for fighting serious or organised crime and tracking down the most dangerous criminals in the UK.
In its notice, the NCA set out details of the facial recognition capabilities it is looking to deploy. Although it has not yet committed to procuring a solution, the notice is the first step in a process which could see it deploy face-identifying technology across Britain.
The document reveals that a facial recognition solution is "upcoming", with its exact abilities set to be decided this year.
Suppliers "have a proven ability" to deliver a variety of capabilities, including 1:1, which means facial recognition which compares a single face against a specific image or database entry to verify an individual’s identity. This method is often used in security applications, such as unlocking a phone, where the system checks if the face presented matches the one on file.
Additionally, the NCA wants to use "1:Many" techniques which compare one face against a large database to identify the person. This is commonly used in settings like airports or security systems, where a captured image is checked against a watchlist or database of individuals.
It is also seeking many:many functionality, which involves the comparison of multiple faces to a large database. This system identifies and compares multiple faces simultaneously, making it ideal for large-scale surveillance or crowd monitoring.
Finally, it requires supplier to have experience delivering retrospective facial recognition (RFR), which is the process of analysing previously captured still images or video footage to identify individuals after an event. This can be performed on single images or in bulk, allowing investigators to match faces to a database to aid in identifying suspects or reviewing historical events.
The provider must "have a proven ability of delivering RFR capability from challenging imagery/video footage" such as "poor quality, off-angle and/or non-frontal" - which means it can identify people even if they are not facing a camera.
Any solution must be "rigorously tested through NIST" and "be able to be used in conjunction, as an automated process, with a large image gallery/data-set held on an NCA-owned IT platform".
It also needs to be "deployable onto an NCA-owned IT platform which has no direct internet access" and capable of being used in conjunction with Officer Initiated Facial Recognition (OIFR), which officers use to identify individuals during investigations or stops, as well as Live Facial Recognition (LFR), which scans faces in public or live footage.
The notice closes with the promise that the database of faces will get larger in the future, making recognition more effective.
"It is anticipated that the facial recognition solution will be used in conjunction with a large image data set," the NCA wrote. "This data-set will be increased and its scope widened over time – as such any solution will need to be able to adapt to any increase in data."
