Skip to main content
Platform / Guard / Shoplifting Detection

Flag the behaviour. Let a human make the call.

Behaviour-based detection of concealment, grab-and-run and till incidents, surfaced for review in real time. QuantumEye packages the evidence and routes it to a manager who confirms before anything happens. Assistive by design, human-reviewed by default.

Illustrative CCTV feed. UK off-licence, watermarked
Why it matters

UK retailers lost £2.2bn to theft in 2023/24 (BRC), 20.4 million theft incidents. QuantumEye delivers real-time AI loss prevention on existing CCTV, ICO-registered and ISO 27001 certified, with human review on every consequential action.

What it does

Three theft patterns. Detected by behaviour, not by item.

The state machine has been tuned on real UK retail footage. Tripped only when the behaviour is unambiguous.

01

Concealment in clothing or bag

Hand-product IoU + hand-bag handoff tracked across 8+ frames. Triggers only when the product is 'lost' past the timeout window.

02

Grab-and-run

High-velocity exit with un-paid product in hand. Multi-camera track follows them to the door.

03

Cash incidents at the till

Unusual till-drawer behaviour, refund without scan, and item swap-out at the cashier.

How it works

A 6-state behaviour model on every camera.

Person state machine: IDLE → BROWSING → PICK → CARRYING → CONCEALMENT → SUSPECTED_THEFT. Tuned for UK retail. Multi-camera handoff via ByteTrack.

EDGE

Per-camera ONNX inference

Runs in the store. No raw video uploads. Detection happens at the camera, so the alert lands before the person reaches the door.

TRACK

Multi-camera handoff

ByteTrack follows the suspect across the store. Cam 04 → Cam 06 → Cam 08, the chain is in the evidence pack automatically.

PACK

Evidence pre-built

Clip, face, confidence score, audit trail and a pre-filled police-report skeleton, on the manager's phone in seconds.

Live example

16:29, alcohol aisle. A bottle goes inside a jacket.

An illustrative scenario from a UK off-licence camera, Zone A. Concealment is flagged on the alert track in real time; multi-camera handoff follows the subject to the exit; the manager intercepts calmly. Item recovered, no confrontation.

app.quantumeye.io/events/shoplifting

Shoplifting

Monitor and review security events in real time

Illustrative
16 May 2026All camerasAll storesPendingConfirmedFalse AlarmAll
Confidence 50% – 100%
Total Detections
9
Last 24 hours
Confirmed Matches
1
100% review accuracy
False Alarms
0
Auto-anonymised
Detection Time
0.4s
Avg edge-to-review
Review queue
Human-in-the-loop · approval required
Concealment in clothing
Pending review
Clip redacted
Live
Match 65%
T-220DB7A657
Location
Aisle 4
Detected
16 May 2026, 12:36
Camera
CAM-04
ConfirmFalse alarm
Concealment in clothing
Pending review
Clip redacted
Live
Match 72%
T-9F1C4E8B02
Location
Aisle 4
Detected
16 May 2026, 12:31
Camera
CAM-04
ConfirmFalse alarm
Concealment in clothing
Confirmed
Clip redacted
Live
Match 73%
T-4A77E2D915
Location
Entrance
Detected
16 May 2026, 11:58
Camera
CAM-01
View report

Pairs well with

GDPR & accuracy

A human reviews every confirmed theft.

QuantumEye flags. A QE Administrator reviews and approves before any action. Alerts default to Pending, never to Approved.

01

Pending by default

Every detected theft is created as Pending. No automated police report, no automated ban. A QE Administrator must approve.

02

False alarms anonymised

If the reviewer marks an event as FalseAlarm, the clip is blurred before it is retained.

03

Append-only audit

Every status change writes an EventAuditLog row. The log is never updated or deleted.

See your aisle, your cameras, your edge cases.

We'll tune the demo to a typical store like yours before the call. A focused 20-minute walk-through, online.