Live facial recognition
Facial Recognition can be used in several ways including
- To support policing in the identification of people wanted for priority offences.
- To support law enforcement including the administration of justice (through arrest of people wanted on warrant or unlawfully at large/ recall to prison).
- To ensure and promote the safeguarding of children and vulnerable people at risk.
Essex Police have been deploying Live Facial Recognition (LFR) technology since summer 2024, following a successful proof of concept exercise conducted in 2023.
During an LFR deployment, cameras are directed at a defined public area. Images of individuals who pass through that area are streamed in real time to the LFR system and automatically compared against a pre authorised watchlist.
If an individual is not on the watchlist, their image is immediately discarded by the system and no biometric data or facial template is retained.
CCTV footage of the deployment area is retained for up to 31 days, in line with policing purposes and standard CCTV retention policies. No other biometric data is stored.
This webpage provides information about our upcoming Live Facial Recognition (LFR) deployments and details of previous deployments, including relevant usage data and system performance.
It also includes the policies and procedures that govern when, where, and why LFR is deployed, helping you understand the legal, operational, and ethical framework within which the technology is used.
In addition, the page contains links to independent scientific and technical reports that outline the testing undertaken, the methodology used, and the results obtained, including assessments of accuracy and performance.
Future deployments
Here you will find dates and locations of upcoming Live Facial Recognition deployments.
Southend, Sunday 14 June 2026
Southend, Friday 19 June 2026
Southend, Saturday 20 June 2026
Southend, Sunday 21 June 2026
Clacton, Sunday 28 June 2026
Previous deployments
Download a CSV file which contains all previous live facial recognition deployment details.
Total alerts
11
Positive alerts
11
Positive interventions
9
Incorrect alerts
0
Incorrect interventions
0
Arrests
0
Other disposals
4
Faces seen
12,566
Incorrect alert rate
0
Watchlist size
2,305
Total alerts
8
Positive alerts
8
Positive interventions
5
Incorrect alerts
0
Incorrect interventions
0
Arrests
1
Other disposals
3
Faces seen
9.469
Incorrect alert rate
0
Watchlist size
2,303
Total alerts
11
Positive alerts
11
Positive interventions
10
Incorrect alerts
0
Incorrect interventions
0
Arrests
4
Other disposals
5
Faces seen
25,570
Incorrect alert rate
0
Watchlist size
2,161
Total alerts
16
Positive alerts
16
Positive interventions
9
Incorrect alerts
0
Incorrect interventions
0
Arrests
6
Other disposals
2
Faces seen
11,776
Incorrect alert rate
0
Watchlist size
2,287
Total alerts
16
Positive alerts
16
Positive interventions
15
Incorrect alerts
0
Incorrect interventions
0
Arrests
8
Other disposals
3
Faces seen
22,951
Incorrect alert rate
0
Watchlist size
2,315
Total alerts
15
Positive alerts
15
Positive interventions
11
Incorrect alerts
0
Incorrect interventions
1
Arrests
4
Other disposals
4
Faces seen
32,976
Incorrect alert rate
0
Watchlist size
2,275
Total alerts
3
Positive alerts
3
Positive interventions
3
Incorrect alerts
0
Incorrect interventions
0
Arrests
2
Other disposals
1
Faces seen
22,101
Incorrect alert rate
0
Watchlist size
2,280
Total alerts
14
Positive alerts
14
Positive interventions
10
Incorrect alerts
0
Incorrect interventions
0
Arrests
2
Other disposals
6
Faces seen
14,626
Incorrect alert rate
0
Watchlist size
2,240
Total alerts
9
Positive alerts
9
Positive interventions
8
Incorrect alerts
0
Incorrect interventions
0
Arrests
4
Other disposals
4
Faces seen
21,862
Incorrect alert rate
0
Watchlist size
2,234
Total alerts
4
Positive alerts
4
Positive interventions
4
Incorrect alerts
0
Incorrect interventions
0
Arrests
1
Other disposals
3
Faces seen
11,703
Incorrect alert rate
0
Watchlist size
2,226
Total alerts
3
Positive alerts
3
Positive interventions
3
Incorrect alerts
0
Incorrect interventions
0
Arrests
0
Other disposals
1
Faces seen
15,549
Incorrect alert rate
0
Watchlist size
1,873
Total alerts
1
Positive alerts
1
Positive interventions
1
Incorrect alerts
0
Incorrect interventions
0
Arrests
0
Other disposals
0
Faces seen
1,000
Incorrect alert rate
0
Watchlist size
2,079
Total alerts
4
Positive alerts
4
Positive interventions
3
Incorrect alerts
0
Incorrect interventions
0
Arrests
2
Other disposals
1
Faces seen
3,965
Incorrect alert rate
0
Watchlist size
2,390
Total alerts
6
Positive alerts
6
Positive interventions
6
Incorrect alerts
0
Incorrect interventions
0
Arrests
4
Other disposals
1
Faces seen
16,308
Incorrect alert rate
0
Watchlist size
1,872
Total alerts
11
Positive alerts
11
Positive interventions
9
Incorrect alerts
0
Incorrect interventions
0
Arrests
2
Other disposals
4
Faces seen
29,612
Incorrect alert rate
0
Watchlist size
2,645
Total alerts
13
Positive alerts
13
Positive interventions
12
Incorrect alerts
0
Incorrect interventions
0
Arrests
4
Other disposals
7
Faces seen
29,310
Incorrect alert rate
0
Watchlist size
2,242
Find out more
General
LFR is technology utilising cameras to scan faces in real time against a pre-determined watchlist to identify people wanted in connection with criminal offences, missing or vulnerable people and to enforce criminal orders.
The algorithm of the software detects whether someone’s face matches that of a person on the pre-determined watchlist.
It if it does, it generates a positive alert. This is only the first stage. Officers then complete a visual comparison and then make an approach to the person to confirm their identity before any further action is taken.
If someone’s face is not matched with that of a person on the pre-determined watchlist, the image and associated data is automatically deleted.
Essex Police are committed to helping people, keeping communities safe, and bringing offenders to justice. Live Facial Recognition (LFR) supports all three objectives.
Police officers spend a significant amount of time trying to locate individuals who are either wanted for criminal offences or at risk of harm. LFR enables this to be done more efficiently and accurately, allowing officers to be redeployed into visible policing and community engagement, where they can have the greatest impact.
A real‑world comparison would be asking an officer to stand on a busy high street with a dossier containing images of 1,000 individuals who are sought by police and requiring them to identify those people as they pass by. LFR simply performs this task in a faster, more reliable, and more proportionate way, using technology to support officers rather than replace them.
Yes, you can see the vans in action during a live deployment and observe the software and technology in use. We publish all deployments on our website.
A positive alert is where the facial recognition technology matches someone on the watchlist from scanning faces. If you are not on the watchlist then the software disregards you within under a second.
A False Positive Alert occurs when the LFR system generates an alert suggesting a match between a live facial image (probe image) and a person on the watchlist, but on operator adjudication or subsequent engagement it is determined that the individual is not the same person as the watchlist subject.
We will always publish the dates and locations of a deployment several days in advance, other than in very exceptional operational circumstances. They will be posted on social media and Essex Police web page.
Data and privacy
If you are not on a watchlist we do not store your biometric data. It is immediately and automatically deleted within under a second and your image is blurred on the screen within the LFR vans. All personal data from the software and watchlists from each deployment is deleted as soon as practicable or any case within 24hours other than details of those subject to a positive alert. If any data is retained for evaluation purposes this would be appropriately documented in line with Data Protection legislation. That data would only ever been retained for the period of evaluation only.
We do retain the CCTV footage for up to 31 days following a deployment. We do obviously retain details of those who are subjected to a positive alert, but this is deleted from the LFR software itself following the deployment. We do retain the details of the positive interventions for further progression and or for analysis. We delete the watchlists within 24 hours of a deployment.
The use of live facial recognition technology by Essex Police is designed to be responsible, proportionate, and fair. It aims to keep the public safe, identifying serious offenders and protecting the vulnerable. Essex Police uses transparency that demonstrates effectiveness, proportionality and compliance with legislation and guidelines when deploying live facial recognition technology.
We will never pass biometric data to third party agencies.
All CCTV footage generated from a mobile CCTV deployment is deleted within 31 days, in the following examples when it is retained:
In accordance with the Data Protection Act 2018, MOPI and the Criminal Procedures and Investigations Act 1996; and /or
In accordance with Essex Police’s complaints / conduct investigation policies.
Watchlists and deployments
There are three operational settings to assist decision making when we deploy LFR.
1. Planned Events such as a football match or festival.
2. General Operational Deployments (based upon ‘legitimate policing requirements’ being met)
3. Dynamic Operational Deployments (e.g. response to significant incident)
All of the above still require justification and this comes in the form of intelligence or crime data to support the deployment. For more detail see the Force policy and procedures on the web page.
A watchlist contains the images and data of those that the Police seek to locate. This includes those who are wanted for crime, subject to court orders, pose a risk to the public, vulnerable or at risk of harm. Unless you are on the watchlist you cannot be matched. When the technology finds a possible match, an alert is generated.
When an officer or team proposes the use of Live Facial Recognition (LFR), they are required to submit a formal application for review and approval by a Senior Officer. This application must clearly set out the lawful policing purpose for the deployment and include supporting intelligence or evidence to justify the inclusion of each individual on the watchlist.
Watchlists are constructed in line with necessity and proportionality principles. The seriousness of the offence directly influences the size and scope of the watchlist: more serious offences may justify the inclusion of a wider group of suspects, whereas for less serious offences, watchlists are typically geographically limited to the town or city where the LFR deployment is taking place.
This approach ensures that the use of LFR is targeted, proportionate, and intelligence‑led, with appropriate oversight at every stage. Further information is available in the Essex Police Live Facial Recognition policy and the associated procedures published on our website.
The LFR alert is only the first stage in the process. 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 watchlist is unique to that event where the technology is being deployed. Only persons who are wanted or suspects will be on an authorised watchlist, and they are unable to opt out. If you wish to physically avoid a deployment this is not grounds on its own for us to have any interaction with you.
Demographic performance and threshold setting
Historically there have been issues with Facial Recognition Technology and potential gender and ethnic bias. As the technology has developed over time this bias has reduced greatly.
Following work by Essex Police with the University of Cambridge and the National Physical Laboratory (NPL), reports has been produced which gives an impartial, scientifically underpinned and evidence-based analysis of the performance of the facial recognition algorithm currently used by Essex Police. As a result:
There is now a better understanding of the demographic performance of the LFR system. The activity we have progressed gives us reassurance that there are algorithm settings at which the system can be operated safely, lawfully and equitably, with no statistically significant difference in performance between demographic groups.
The accuracy of the LFR algorithm is informed be the ‘sensitivity threshold’, the sensitivity threshold setting allows for the software to be adjusted to try and identify as many people as possible on the watchlist balanced against reducing the risk of falsely identifying members of the public. When the threshold setting is adjusted it can have an impact on the level of disparity between certain groups (ethnicity, age or gender). The basic principle is that if the threshold setting is raised then you are less likely to identify as many people as possible but will reduce the risk of false positives. Lowering the threshold increases identification but also raises the risk of false positives, while potentially reducing bias in positive identifications. Once the threshold level has been set then the software will only alert to potential identifications above that score.
Essex Police has deployed LFR at a sensitivity threshold of 55 since August 2024, achieving strong operational results as part of its commitment to fighting crime and keeping communities safe. To ensure continued compliance with legal obligations under the Public Sector Equality Duty, Essex Police, in conjunction with the Home Office and the National Police Chiefs’ Council (NPCC), commissioned two independent studies to evaluate the performance of the algorithm used by the force. These studies were conducted by the University of Cambridge and the National Physical Laboratory (NPL).
Both studies found differences in how well the algorithm identified certain demographics, with Black males being recognised most effectively. The Cambridge study reported this disparity as statistically significant and suggested lowering the sensitivity threshold below 55. However, the NPL study did not consider the disparity significant, stating that the algorithm was equitable at sensitivity thresholds of 55, 60, and 65 in the True Positive Identification Rate and False Positive Identification Rate.
Upon receipt of the NPL study, Essex Police sought advice from Professor Paul Taylor, the Police Chief Scientific Advisor, to assist with interpreting the results of the two reports. This advice has supported our decision making with regards to the 55 sensitivity threshold that we will seek to deploy at, which based on all the findings and advice represents an equitable setting with minimal risk of false positives.
What this means is that alerts are only generated where the similarity score meets or exceeds 55. Independent scientific testing indicates that this threshold achieves equitable performance across different demographic groups.
Statistical significance quantifies whether the observed performance difference is likely due to chance, or due to some underlying factor. Following convention, a 0.05 significance level was set prior to evaluation and analysis of results.
The risk of a member of the public being falsely identified through the use of Live Facial Recognition (LFR) is very low.
To date, Essex Police have recorded four false positive alerts across 76 deployments and more than 108,000 faces scanned. In each case, the false positive alerts were attributable to poor‑quality images being ingested into the system, rather than failures of the live matching process itself. Following identification of this issue, Essex Police worked directly with the software provider to recalibrate the system, ensuring that images of insufficient quality are now automatically rejected.
Neither of the independent academic studies commissioned by Essex Police identified evidence of a systemic or significant bias that would place the public at risk of being falsely identified. At the threshold setting adopted by Essex Police, the likelihood of a false positive alert is less than 0.2%, equating to fewer than one false alert per 57,000 faces scanned.
In addition, LFR is supported by multiple operational safeguards. Any alert generated by the system is subject to a manual visual review by a trained officer before any engagement takes place. This layered approach ensures that technology supports, rather than replaces, human judgement and further reduces the risk of inappropriate police interaction
Live Facial Recognition Documents
Report 1
Live Facial Recognition: Accuracy, Watchlists and Deterrence
Live Facial Recognition: Accuracy, Watchlists, and Deterrence. PDF 5Mb
Dr Matt Bland & Jacob Verrey, University of Cambridge. Version 6. 12 March 2026.
Report 2
Accuracy and Equitability Evaluation of Corsight Apollo 4 Live Facial Recognition
Accuracy and Equitability Evaluation of Corsight Apollo 4 Live Facial Recognition.
Data Science and AI Department, National Physical Laboratory. March 2026.
Report 3
Live facial recognition: Algorithm Performance, Testing and Threshold Setting
College of policing app for Live Facial Recognition
Science and Technology in Policing – operational testing of facial recognition technology
NPIA – Police standard for still digital image capture
Government Fact sheet on police use of facial recognition
Public Attitudes to police use of facial recognition technology
Thomson & Carlo v Metropolitan Police Commissioner summary
Bridges v Chief Constable of South Wales Police summary
Essex Data Ethics minutes dated 20th November 2025