Live Facial Recognition Technology

All LFR deployments use a CCTV feed to capture facial images before they are input into the LFR system for the purpose of matching. CCTV footage created for this purpose is recorded and kept for up to 31 days except in some very rare instances where it needs to be retained for evidence. This is in line with most other public CCTV systems.

If a captured face is not on a watchlist (i.e. the system does not match the face with one held on the database) we will not store this photograph or facial template (biometric data). They are immediately and automatically deleted after being captured.

Where a match is made, we will delete the photograph and the facial profile (biometric data) relating to it as soon as possible, after it has been used. This is usually during or at the conclusion of the LFR operation, but in all cases it is no longer than 31 days after the conclusion of the LFR operation.

Each watch list is created specifically for a single deployment of the technology, and will contain photographs and details of people who are wanted by the police for committing a crime or breaching conditions such as court orders, or wanted by the court or prisons. In some exceptional circumstances, watch lists may include photographs and details of persons who are missing and believed to be at risk of serious harm. Unless you are on the watch list, you cannot be matched. South Wales Police have led in the deployment of this technology, and in their deployments there has not been a single wrongful arrest because of their use of facial recognition.

The use of live facial recognition technology by Thames Valley Police is designed to be lawful, responsible, proportionate, and fair. It aims to keep the public safe, identifying offenders and protecting the vulnerable. Thames Valley Police aims to be as transparent as possible in our use of the technology in order to demonstrate effectiveness, proportionality and compliance with legislation and guidelines when deploying live facial recognition technology. We will always publish the dates and locations of a deployment several days in advance, except in very exceptional circumstances, and will always publish the results of those deployments.

Only persons who are of specific interest to the police will be on an authorised watch list, and nobody is able to opt out of this.

If you wish to physically avoid the technology, you may choose to do so, and you will not be stopped or obstructed. Avoiding the technology will not give a reason, on its own, for officers to have any interaction with you. We will always publish the dates and locations of a deployment several days in advance, except in very exceptional circumstances.

When the technology finds a possible match, an alert is generated. Any alerts are verified by an operator prior to an intervention by officers on the ground. The system will only seek to match those individuals placed on the watch list. Each watch list is unique to that event where the technology is being deployed.

No. The percentage can vary and is impacted by several factors including the scope of the watch list and how many people pass through the zone of recognition for the camera. Both of these vary considerably in respect of each unique operational deployment.

Thames Valley Police will do everything reasonably possible to inform the public where and when we intend to use Facial Recognition technology, for example:

• Online before it is used (including the results after LFR is used),
• Information leaflets,
• Posters and signs,
• Police officers who will be available at all deployments to explain how LFR works and why we are using it.

Historically, there have been some issues identified in relation to potential gender and ethnicity bias in respect of some types of Facial Recognition Technology. As the technology has developed over time this bias has reduced considerably.

The National Physical Laboratory has tested both the algorithm used by TVP for Live Facial Recognition (NEC Neoface) and the Home Office’s Police National Database (PND) algorithm used by TVP for Retrospective Facial Recognition (Cognitec FaceVACS DBScan ID v5.5).

Reports have been produced which give impartial, scientifically underpinned and evidence-based analysis of the performance of the algorithms, and the key findings show that:

• Live Facial Recognition: At the higher thresholds used by TVP, false positives are zero, and no demographic bias is detected. Read the full report:
  Facial Recognition Technology in Law Enforcement Equitability Study Final Report,
• Retrospective Facial Recognition (PND): At the settings used by TVP, if a correct match is on the database, the system will find it in 99% of searches. However, in a limited set of circumstances the algorithm is also more likely to incorrectly include some demographic groups (Black, Asian, female and younger people) in its search results. Read the full report:
  Facial Recognition Technologies: Accuracy and Equitability Evaluation of Cognitec FACEVACS-DBSCAN ID V5.5

In order to mitigate against the risks presented by 'false-positive' results in its retrospective facial recognition searches, Thames Valley Police has implemented a number of measures around its use of the technology, in line with national guidance. Importantly, every facial match is reviewed by a human to ensure the match is accurate in their professional judgement, and no result is ever used as evidence – it is simply an intelligence indicator that the same person appears to feature in both images.

The use of Live Facial Recognition (LFR) and Retrospective Facial Recognition (RFR) by TVP have been subject to an Equality Impact Assessment (EIA). Read the LFR EIA. The RFR EIA is due to be published shortly.

If you have complaints, questions or comments about Thames Valley Police's use of Live Facial Recognition, please contact us at [email protected]