CarMen aims to develop a fully automated, scalable biometric border control system capable of continuously identifying both pedestrians and passengers in moving vehicles —cars and trucks— “without stopping the vehicle, opening its windows, or getting out” of it.
Meant for both land and sea borders, the project is positioned as nothing short of transformational: “The CarMen project is set to revolutionize the way borders are managed across Europe”, reads its official website, playing “a crucial role in shaping the future of border management”.
This “revolution” rests on a radical ambition: enabling real-time biometric identification through DTCs (Digital Travel Credentials, stored on a mobile application on the traveller’s smartphone) and facial recognition in vehicles moving up to 20 km/h, in both day and night, across indoor and outdoor settings, and under variable weather conditions.
The stated performance target is as ambitious. According to a document — the project’s Grant Agreement — that we obtained in redacted form from the Research Executive Agency, these include the detection of “90% of faces in slowly moving vehicles” and having “at least 80% of vehicles correctly checked automatically using the biometrics in their DTC.”
Through this innovative system based on sophisticated and heterogeneous “biometrics on the move” — something that “has been required by Frontex as early as 2017” — “up to 4 passengers including the driver” will be detected in “slowly moving vehicles”, reducing “the border control processing time by 25 to 50% for bona fide travellers”. The system will also allow border guards to check passengers in coaches “walking through the coach, using dedicated portable equipment as they walk”, and help avoid congestions at ports, where customers will only be required to arrive 30 minutes before departure, rather than 1 hour, “which is the actual time required”, reads the document.
Expected impact is also ambitious, aiming at a reduction of waiting times “by up to 90%”, and promising the ability to “identify potential threats in real-time” through a “seamless and non-intrusive border control process”. This should even “support environmental sustainability by reducing the need for physical infrastructure expansions, in line with the European Green Deal”. Consequently, “frugal AI approaches” will be adopted to limit energy consumption.
Crucially, travellers will need “to have their smartphone switched on and the mobile application launched” while approaching to cross the border.
According to deliverable D1-D14.5 ‘AI-Requirement No.5’, which was also disclosed to us by the REA, CarMen’s objectives include the detection of “anomalies in biometrics on-the-move using behaviour analysis”. This means “allowing the system to have cues, even before using physical biometrics, about the potential intent of pedestrians and attempt to detect movements that look suspicious; analysing movements of the body that could be a sign of nervousness, that look suspicious, and so on, to provide a global evaluation of the confidence level attributed to a traveller,” reads the document.
CarMen’s Grant Agreement highlights that such a system is currently “not operational anywhere in Europe”, and that the project is “made at the same time with the close and more distant future in mind”.
A video presentation on the project’s website describe how the CarMen framework would work once operational.
CarMen will develop several innovative solutions for “biometrics on the move”:
1) A new form of iris recognition on-the-move that works “in near infrared light with a focus on iris templates extracted cameras captured by body-worn cameras, to assess image quality and o perform iris recognition on the move”
2) “FACE & Periocular Recognition” focussed on
a) the problem of face detection “through cars’ windows (lateral windows as well as windshield), and their suitability for face recognition against ID document face data”;
b) “face recognition robust to pose, degradations and motions (…) specific to on the move conditions”
c) robust periocular recognition “based on image regions extracted from high resolution acquired images from slow moving vehicles”.
“Periocular” recognition is defined, in the document we obtained, as “one of the promising biometric traits for human recognition” as “It encompasses a surrounding area of eyes that includes eyebrows, eyelids, eyelashes, eye-folds, (…) eye shape, and skin texture”. Based on machine learning and deep learning techniques, “its relevance was emphasized during the COVID-19 pandemic due to masked faces”.
3) A new “Heterogeneous Face Recognition” system will also be researched, one “which is capable of processing and matching facial images from diverse source and modalities”. “This is crucial for matching with face images from passports”, reads the document.
Thanks to these promised technological developments, traveller authentication will be “seamlessly” performed via smartphone through Digital Travel Credentials (DTCs), in two steps:
“1) the biometric data (face, iris, periocular) of travellers is securely stored in their smartphones, thanks to a DTC (Digital Traveller Credential); 2) the biometric data is securely transferred from the DTC to the border authority infrastructure, and compared to the live biometric data collected as the traveller crosses the border control point.”
Physical documents will not be needed, according to CarMen’s vision.
The project “enables the use of fix and body-worn cameras”, but also envisions “Behavioural biometric analysis”. In fact, according to the Annex to the document we obtained, “in some cases feeds from closed-circuit cameras will be exploited to infer the trajectories of pedestrians — and therefore anticipate sudden worrying behaviours”.
Once travellers arrive at the identification system, besides automated behavioral analysis, another layer of automation will intervene. Reads the Grant Agreement: “At this stage, the device used at the first step to pre-enrol (the DTC stored on the traveller’s smartphone, ndr) will connect seamlessly to the identification system through an unwired connection, and without action from the traveller, to provide the biographic and biometric information that could allow for the identification.”
Incidentally, “the identification system will also be usable for health problem detection”.
A crucial table, contained in the redacted Grant Agreement we obtained, illustrates “National and international related research and innovation activities” to CarMen.
Several of the projects we are investigating are featured. Among them: SMILE, IMARS, D4FLY, FLEXI-cross. In addition, some more recent projects in border security are also mentioned, such as SafeTravellers (2024-2027) and EINSTEIN (2024-2027). The inclusion of the French SafeCity project is noticeable as well, as it involved a “complete security system” for multiple purposes, from road safety to school security, to be developed by several CarMen project participants: Thales (project leader), Idemia, Arclan and Yncréa.
Interestingly, all of the above-mentioned projects involve(d)“CarMen participants”.
From the individual contributions of each past project, we can reverse-engineer which components CarMen will implement/build on. Namely:
a) “a specific architecture to accelerate border crossing” from SMILE (H2020)
b) Morphing Attack Techniques to fight fraud in travel documents (crucial to CarMen as “passport validity must be checked before authorizing any sort of DTC”) developed during project iMARS (H02020)
c) techniques to identify travellers in vehicles developed during project PROTECT
d) UREAD’s (University of Reading) experience developed as a partner of D4FLY, in particular on “enrolment (kiosk based) and on-the-move-verification (including biometric fusion)”, together with an evaluation of the notion of “biometric corridor” equipped with 2D face, 3d face and iris recognition
e) KEMEA’s experience as leader of the Greek trial of the FLEXI-cross project, which developed, deployed and validated “a toolkit of innovative border-checking solutions, in real operational environments, addressing road, rail and port borders”.
A difference is however noted between CarMen and previous biometric solutions for vehicles at the borders from EU-funded projects, such as PROTECT’s and FASTPASS’s: a) “cars had to stop and an ABC gate-like equipment had to be moved in front of each passenger window” (a process deemed too slow); and 2) these legacy solutions were relying on traditional passport readers, whereas all functionalities of CarMen will be available on the passengers’ mobile phones.
Interestingly, the forthcoming EU Entry-Exit System is considered a “Threat”, “as we do not know how this will impact the land and sea borders”.
A joint event held in July 2025, and featuring projects CarMen and BorderForce, also illustrates how the biometrics on-the-move technologies developed within the former are meant to integrate with the “self-sufficient”, “transportable”, real-time surveillance assemblage prototyped by the latter, and are therefore “complementary”: “Together, these projects form two halves of a holistic approach”, reads a blogpost: “CarMen brings the intelligence, BorderForce delivers the oversight”.
Three use cases are envisioned, based on the needs of end users (“border police and transport operators”): for 1. vehicles and goods, 2. pedestrians, 3. coaches. Results of these use cases “will be integrated” into further project tasks, “providing a crucial real-world context to support the technical initiatives” in other working packages.
The document we obtained from REA gives some more detail on the three envisioned use cases:
a) pedestrians will be channeled in “streams”, and the detection system will be efficient even if subjects “do not look directly at the cameras”; “anonymous behavioural analysis, thanks to video cameras not specific for the project (closed-circuit cameras for example)” will also “detect suspicious movements of pedestrians who may try to find a way to fool or to bypass the identification system”
b) for cars and trucks
c) for coaches and buses (“it will be faster to keep the travellers inside the coach, instead of getting them off for identification. Then, a controller will get on the coach to identify people with its body-worn camera, instead of having people getting in front of the sensors”.)
In terms of project pilots, “Carmen solutions will be demonstrated in operational conditions over the UK – France border, as French and UK border authorities are part of Carmen consortium, using the infrastructure proposed by the partner Brittany Ferries.”
More precisely, CarMen will be tested in three different locations, in Greece, France and the UK –various pilot sites to are meant to “ensure the robustness and scalability of the biometric systems across different border environments”.
Field demonstrations and testing will also be “specifically designed for identifying individuals inside moving vehicles, including both cars and trucks”.
Crucially, “By addressing the unique challenges of identifying individuals in moving vehicles, this task ensures that the CarMen system is adept at handling real-world scenarios at border checkpoints, ensuring seamless flow and accurate biometric identification”, project participants promise.
Testing conditions will approach real-life scenarios gradually: “Even though a preliminary round of test will happen at project mid-term in the labs, the operational conditions that will be found in the final test will be reproduced as much as possible in the lab”.
In fact, “A final test” will be performed in Le Havre, “at a real border place”. Also, “In its final phase, this research will engage citizens from multiple countries in realistic border-crossing scenarios”, reads the description for D14.6 – OEI – Requirement No.6.
In terms of exploitation, the strategy will “cover a 2-to-5-year period beyond the project’s closure” — and in particular, “18 months beyond the project’s closure, marketing strategies will be initiated to promote the developed tools”.
Even if the technological solutions are new, it does look like coming as close to actual deployment as possible is the goal here.
More pragmatically the idea, in terms of Outcomes, is to assess the solution once tested: “From test in operational conditions of an innovative face biometrics approach, decide if we move forward with a “non-stop biometric vehicle control at the border” new product (provided the market study also shows potential revenues)”, reads the project’s Grant Agreement.
In order to be actually deployed, the system will not only have to satisfy the needs of border guards, but also be able to generate revenues.
While all other Working Packages enjoy detailed descriptions in the Grant Agreement, the one dedicated to ‘Ethics requirements’ merely states the self-evident claim that “this work package sets out the ‘ethics requirements’ that the project must comply with”.
And yet, when it comes to the result of the ethics assessment procedure, we can read that the project only received “conditional clearance”: “The proposal has been subject to an Ethics Screening and the outcome of the Ethics Screening is: Conditional clearance. The Post-Grant Requirements specified in the EthSR will become contractual obligations. The Post-Grant requirements have been transferred automatically into 6 ethics deliverables. An extra work package on ethics has been automatically created by the system.”
Also, “Expertise has been sought outside the consortium” (but resumes of the experts are fully redacted in the document we obtained, to protect their privacy).
The approach to AI ethics is however pretty sophisticated in general throughout the Annex to the GA, usually mentioning strategies of intervention and guidelines. This means that the project develops technological solutions that are inherently dangerous, at least to some degree.
CarMen however plans to enjoy the same structural opacity that most EU-funded border security projects enjoy. Crucially, the “List of Deliverables” contained in the Grant Agreement reveals that all but three of the official documents that the project will produce over its lifespan will be labelled “SENSITIVE”, i.e. “Limited under the conditions of the Grant Agreement”.
This means that just 3 out of 36 deliverables will be public.
These are:
D1.2 – Data Management Plan, legal and ethics requirements
D12.4 – Lessons learned register
D13.2 – Lessons learned register bis.
Interestingly, a whole deliverable on Ethics will be dedicated to AI. Also, a Human Rights Impact Assessment is expected, covering “development, deployment and post-deployment phases”. The ALTAI framework for self-assessment is referenced.
Even though the Consortium openly admits that “the research incorporates high-risk artificial intelligence applications for biometric identification, involving collaboration with state authorities”, the AI Act is not mentioned in several ethics-related passages.
Stunningly, we can even read that
“Relevant aspects connected to the development, deployment and implementation of AI tools connected to their performances in relation to human rights impact (including on potential vulnerable groups of people) have not been adequately taken into consideration.”
A further passage from the Grant Agreement (Annex 1, part B) reveals that an extremely dangerous idea — that of applying biometrics at the border NO MATTER its risks, the unknown consequences, and NO MATTER what even EU law says, or the impact on and feedback of citizens — is held by several governments and private entities:
“The systematic application of biometrics to border control is in any case currently advocated by various governmental and private institutions. However, the societal implications of this measure have yet to be properly assessed at a time when AI-based technologies are evolving at a rapid pace, and while stakeholders may no longer have enough time to consider the impacts on citizens and what is considered acceptable”.
A rational and humane approach would suggest not to develop and/or deploy such technologies until we have a better better idea of their effects, whether they are desirable/acceptable to the public, and whether they can be reconciled with democracy. Instead, this becomes marginal consideration in a section concerning the “Integration of social sciences and humanities” for a project that aims to actually normalize mass use of biometrics (on the move) at EU borders.
