Platform / Digital Video Analytics / Vehicle Insight System
• REC   CAM-VLA-07 · MAIN ST / DOCK ROAD · 2026-05-03 14:03:22Z LAT 13.8506° N · LON 60.9833° W · 4 SUBJECTS · 1 FLAG

DIGITAL VIDEO ANALYTICS · VEHICLE INSIGHT SYSTEM

Every vehicle in frame, resolved.

Vehicle Insight System turns the cameras you already operate into a sovereign automatic number plate recognition and vehicle intelligence network. Read every plate, classify every vehicle, reconcile against registration and ownership records, and escalate the ones that don’t match, in under a quarter second.

~240 msDetect to score
4 / frameConcurrent subjects
97.3%Plate read accuracy

PIPELINE · 04 STAGES

From frame to enforcement, in four stages.

Vehicle Insight System is a four-stage pipeline that runs on every frame of every camera you point at it. The stages are sequential; the latency is end-to-end ~240 ms.

STEP 01 / 04

Ingest any feed

ONVIF, RTSP, NVR replay, body-cam, drone, or fixed traffic camera. No specialist hardware. Existing CCTV is enough.

STEP 02 / 04

Detect & classify

Per-frame detection of vehicles by class, make, model, generation, body colour, and direction of travel.

STEP 03 / 04

Read plate

Multi-jurisdiction plate OCR with confidence, glare correction, and partial-match recovery for occluded characters.

STEP 04 / 04

Reconcile & score

Cross-checked against DVLA registration, NEXUS watchlists, and SENTINEL risk model. Mismatches escalate without an officer in the loop.

RECORD STRUCTURE

What the model extracts, per vehicle, per frame.

The vehicle to the right is observation 02 in the hero frame above. Tag does not match VIN. SENTINEL escalated. The case file writes itself.

RECORD · VEHICLE 02

READ AT 14:03:22Z · CAM-VLA-07 OCR 0.96
PCN 4421
ISSUING AUTHORITY    LCA · ST LUCIA DVLA
REGISTERED PLATE      PCN 4421
REGISTERED VEHICLE    TOYOTA RAV4 (MISMATCH)
OBSERVED VEHICLE      NISSAN FRONTIER · GREY

Every observed vehicle generates a structured record. The record below is the canonical shape: every field is populated per detection, every detection is bound to a frame, every frame is bound to a camera and a timestamp.

MakeNissan
ModelFrontier (D40)
Body classPickup · 4-door
ColourGrey · #6f7681
PlatePCN 4421
Speed0 km/h
HeadingN · 4° dev.
LaneL2 · NB
Owner of recordRental Co. — Avis LCA
Registration statusTag does not match VIN
NEXUS hits2 · adverse media
SENTINEL score0.81 · escalate

CAPABILITIES · 09

Nine primitives. One vehicle intelligence engine.

Each primitive is independently deployable and policy-governed per camera. Combined, they form a comprehensive vehicle intelligence stack engineered for sovereign government deployment.

01 / 09

Multi-jurisdiction ANPR

Plate recognition across OECS, CARICOM, US, EU, and UK formats. Stacked, two-line, and motorcycle plates supported. Glare and motion blur recovered, not failed.

02 / 09

Make · model · colour ID

Vehicle classification down to model generation. Colour assigned in CIE-Lab so a grey RAV4 at dusk and a grey RAV4 at noon resolve to the same record.

03 / 09

Tag-to-VIN reconciliation

Every read is compared against DVLA registration. The plate, the make, and the colour must agree. They don’t always, and that’s the point.

04 / 09

Cross-camera tracking

Persistent vehicle ID across cameras, junctions, and zones. One subject, one record, one timeline, even after the vehicle leaves frame for ten minutes.

05 / 09

Speed & lane analytics

Per-subject speed estimation from monocular video, lane discipline scoring, illegal turn detection, and stop-line compliance, with no inductive loops to install.

06 / 09

Convoy & follow-me detection

Two or more vehicles travelling in formation across multiple cameras are surfaced as a convoy event. Useful for trafficking flows, motorcades, and tail surveillance.

07 / 09

Watchlist & BOLO escalation

BOLO plates, stolen tags, and NEXUS adverse-media hits raise officer alerts at the camera nearest the subject, not at a control room two cities away.

08 / 09

Geofenced zones

Per-junction, per-lane, per-shift policy. Rules apply to the camera, not the operator. Audit is automatic.

09 / 09

Forensic search

Find every silver pickup with a partial plate ‘…CN 44…’ that crossed Main St between 13:00 and 15:00 last Tuesday. Retrieved in seconds, not days.

OPERATOR CONSOLE · LIVE

What a shift looks like.

A representative eight-read window from a vehicle ops console: three flags, one BOLO hit, one obscured plate. All vehicles and plates shown are fixtures.

• REC   OPERATOR CONSOLE · OPS-02 · 14:03:22Z 36 cameras · 8 reads · 3 flagged · 1 BOLO
TIMECAMPLATEOBSERVEDCOLOUROCRSTATUS
14:03:22ZVLA-07PCN 4421Nissan FrontierGrey0.96TAG MISMATCH
14:03:22ZVLA-07JCN 1234Toyota HiluxBlue0.98MATCH
14:03:22ZVLA-07BAR 5678Mazda 323White0.91UNVERIFIED
14:03:22ZVLA-07SVG 9012Mitsubishi L300Red0.94MATCH
14:02:58ZVLA-12LCA 0871Hyundai TucsonBlack0.88INCONCLUSIVE
14:02:41ZVLA-03GD-221-KSuzuki JimnyGreen0.97MATCH
14:02:18ZVLA-19Honda FitSilver0.42PLATE OBSCURED
14:01:55ZVLA-07JM 88401Ford RangerWhite0.93BOLO HIT

DEPLOYMENTS

Where Vehicle Insight System earns its place.

Four operational patterns we deploy alongside Border Intelligence, NEXUS, and SENTINEL. All numbers are illustrative of in-production deployments under jurisdictional NDA.

BORDER & PORT

Vehicles entering and leaving the port

Every vehicle crossing the port gate is read, reconciled against the manifest, and bound to the driver’s biometric capture in BORDER INTELLIGENCE. Smuggling vectors close before the trailer reaches the highway.

+38%Manifest matches
−62%Hand-search workload
POLICE & LAW ENFORCEMENT

BOLO and stolen vehicle recovery

Stolen tags, BOLO subjects, and active warrants resolve at the nearest junction camera within 240 ms of the vehicle entering frame. Officer push alerts go to the unit on patrol, not the desk.

240 msDetect to push
1,600+DBs cross-checked
TRANSPORT & ROADS

Lane discipline and stop-line compliance

Illegal turns, stop-line violations, hard-shoulder running, and bus-lane misuse are detected on existing traffic cameras. No loops, no gantries, just the cameras already on the pole.

0 sensorsRoadside hardware
97.3%Plate accuracy
FORENSIC INVESTIGATIONS

Retrospective vehicle search

Investigators query the full historical archive by partial plate, vehicle attributes, route, or time window. A grey pickup with a plate ending ‘21’ that crossed Main St last Tuesday is one search away.

Sub-secSearch latency
100% auditEvery query logged

SPECIFICATIONS

The numbers, on the record.

Detection

  • Frame rate25 fps · 1080p
  • Min. plate height14 px
  • Plate OCR (clear)97.3%
  • Make/model accuracy94.1%
  • Colour (CIE-Lab)ΔE < 6
  • Detect → score~240 ms

Sources

  • CameraONVIF · RTSP · NVR
  • MobileBody-cam · vehicle dash
  • AerialUAS / drone payloads
  • RecordsDVLA · Customs · NEXUS
  • ISR FusionAIS · ADS-B link-in
  • Hardware add-onNone required

Sovereignty

  • HostingSovereign hybrid cloud
  • ModelsIn-house · no licensed AI
  • RetentionPolicy-bound · default 7d
  • AuditPer-query, immutable
  • Operator separationRBAC · per-junction
  • Time to operational~6 weeks

GOVERNANCE

Vehicle Insight System inherits the VEILWATCH Sovereign Hybrid Cloud Architecture. Plate reads, vehicle imagery, and ownership joins never leave the government’s controlled environment. Cloud compute executes non-invertible payloads; cloud providers never see underlying data. Retention defaults to seven days and is governed by your jurisdiction’s policy, not ours.

SOC-2 TYPE II GDPR ART. 9 FATF ISO 27001 SOVEREIGN HYBRID CLOUD

DEPLOYMENT

Pilot one junction in week one. Cover the country in week six.

Vehicle Insight System runs on the cameras you already own. Bring one feed, one registration database, and one operator. We bring everything else.