Computer Vision For Defence And Law Enforcement
Processing inside your boundary, with no outbound connection and no vendor telemetry. Deployed across defence installations and national police operations.
Core Applications
Face Recognition, Vehicle Forensics And Perimeter Security
Defence and police sites already have cameras on every gate and fence line, and the guards cannot watch every screen at once. Access cards get lent, plates get swapped, a VIN is ground off and restamped, and a fence is crossed in the small hours on a feed nobody is looking at. Video analytics checks each face at the gate, reads each plate and flags each crossing inside your own boundary, and keeps a record of every event for the officer who has to answer for it.
- Gate camera
- Server inside your boundary
- Alert to the control room
Biometrics & Liveness
Face recognition at gate and perimeter, 1:1 verification and 1:N identification. Active and passive liveness checks to defend against spoofing and replay attempts.
Vehicle Forensics
Licence plate recognition and VIN tamper detection for forensic verification. Vehicle counting and classification at checkpoints across the network.
Perimeter Security
Restricted-area entry monitoring and loitering detection. Triggers instant alerts when unauthorized individuals cross virtual tripwires near secure installations.
Portfolio
Deployment References
Restricted Area Protection System
Ministry of Defense, Republic of Indonesia
Facial recognition perimeter security deployed at Indonesia’s Ministry of Defense installations, combining millisecond biometric authentication with threat detection.
License Plate Recognition & VIN Tamper Detection
National Police, Republic of Indonesia
Licence plate recognition and VIN tamper detection for physical vehicle registration checks. Manual inspection replaced with automated forensic verification.
“A reliable partner with solid technical capabilities, particularly in AI CCTV with Face Recognition and restricted-area protection.”
Denny Fitrianto
Official, Cyber Defense Center, Ministry of Defense, Republic of Indonesia
Challenges
What A Secure Site Deployment Has To Meet
Air-Gap And No Telemetry
No outbound call at inference time. Licence activation and model updates delivered as signed offline packages, transferred by your own approved media process.
Procurement Structure
Subscription commitments are difficult under most capital budget rules. Perpetual licensing with contracted support is the default structure for this sector.
Audit And Chain Of Custody
Every match, every threshold change, and every access event is logged. What ARSA does not do: produce forensic or evidentiary determinations. The system returns evidence and confidence scores; the determination is yours.
Camera Placement For Face Recognition And Perimeter Detection
Each analytic needs its target large enough in the frame and seen from a usable angle. Our camera standard, ARSA-CVS-001, sets the figures every camera on a secure site is checked against:
- Cameras: 2 MP minimum, 4 MP recommended, at 15 fps or more.
- Face matching at a gate: 1:N watchlist matching needs at least 533 PPM (pixels per metre) at the face, a tilt of 5 to 15 degrees and never steeper than 15, and at least 100 lux. A 2 MP camera holds that density to about 3.1 m and a 4 MP camera to about 4.2 m, so gate cameras sit close to the lane and nearly level with the face.
- Perimeter and restricted zones: intrusion detection needs 71 PPM, a tilt of 15 to 40 degrees and no more than 55, and 10 lux at the target; a 2 MP camera covers about 23.6 m and a 4 MP camera about 31.4 m.
- Plates at checkpoints: licence plate recognition needs 325 PPM and a tilt of 10 to 25 degrees, 30 at most, which keeps a 2 MP camera within about 5.1 m of the plate.
- Outdoor gates and fences: IP66 housings, with IK08 recommended, and clocks synchronised by NTP to under one second, which cross-camera correlation and evidential export depend on.
Maximum distance with a standard 60 degree lens
One-to-one verification at a desk or turnstile and VIN checks at an inspection bay are specified position by position during the assessment. If one of your cameras falls short, the on-site assessment says which one and how to move it; for non-classified footage that can be shared, the free AI assessment does the same from three clips. The full figures are on the system requirements page.
FAQ
AI For Government FAQ
Can the system run fully air-gapped?
Yes. No outbound connection at inference time and no telemetry. Licence activation and model updates are delivered as signed offline packages.
Is source code or are model weights transferred?
No, under any licence tier. Source escrow is available under separate agreement.
Can we buy perpetually rather than by subscription?
Yes, and for this sector it is the default. Perpetual licensing with a contracted support term.
Is output admissible as forensic evidence?
No. ARSA returns confidence scores and logged evidence, and the determination stays with your investigators. Evidentiary use requires your own process and, in most jurisdictions, human verification.
What accuracy can we expect at our gate positions?
Benchmark figures are laboratory results. Deployment accuracy depends on camera position, pixels between the eyes at the detection point, and lighting. A feasibility assessment establishes it against your actual positions.
Should government face recognition run on-premise or in the cloud?
On-premise, for most agencies. Face templates are biometric data: matching inside your own boundary keeps the vendor out of the data path, keeps the system working when external links are down, and allows a fully air-gapped installation. The on-premise face recognition and liveness SDK runs entirely on your servers, with 1:1 verification, 1:N identification and liveness, and is available as a perpetual licence for capital budgets.
How do we test AI on cameras whose footage cannot leave the site?
Request an on-site assessment. ARSA engineers check each camera against ARSA-CVS-001 inside your boundary and measure the models on your actual gate and perimeter positions. For non-classified footage that can be shared, the free AI assessment returns an annotated clip, a report and a 30-minute call within 3 business days.
Does the system keep an audit log of matches and operator actions?
Yes. Every match, every threshold change and every access event is logged on your own servers, with the time and camera of each event, and the log can be exported for your internal audit. Whether a process meets your agency’s rules stays with your auditor; ARSA supplies the records.
Can it read plates and verify VINs at a vehicle checkpoint?
Yes. Licence plate recognition reads plates at gates and checkpoints, and VIN tamper detection supports physical vehicle registration checks, as in the vehicle registration platform ARSA built for the National Police. Each read is logged for your officers to review.
Talk With Our Consultants
Footage from defence and law enforcement sites usually cannot leave the building, so the assessment comes to you. ARSA engineers review your camera positions, lighting and network inside your boundary and measure our models against your actual gate and perimeter positions, and the assessment fee is deducted from the project. Where you hold non-classified footage that can be shared, you can also send us up to three one-minute clips for a free AI assessment.
Every deployment follows the same steps: a site survey, a check of each camera against our placement standard, installation on an AI Box or on your own servers, tuning on your footage and a handover. Processing stays inside your boundary and works without an internet connection, and the clips you send for an assessment are never used to train our models. ARSA built the restricted-area security system for the Ministry of Defense.
Related reading: air-gapped AI deployment for defence and government sites, why security leaders keep video analytics on premise, edge AI and data residency rules and ANPR gate access control.