INDUSTRY WORKFLOW

AI CCTV evaluation for warehouses

Assess loading bays, stock areas and after-hours approaches against the way your warehouse actually operates. Delivery activity and legitimate work should be part of the test.

Start with the operating workflow

Warehouse supervisors coordinate deliveries, cleaning and stock movement. Security needs the zone, hours and exceptions so an alert is actionable rather than simply another notification.

Define an observable event

A person enters a closed loading bay after dispatch, excluding a scheduled collection.

Limitations to test

Forklifts, pallets, vehicle occlusion and open doors can change a scene substantially. Detection is not a stock ledger and does not prove inventory shrinkage or theft prevention.

Technical requirements

  • Camera views with usable target detail beyond racking and parked vehicles.
  • Day/night testing under actual bay lighting and vehicle headlights.
  • Delivery schedules, network/processing and named shift response owners.
Camera and stream requirements →
IllustrationNot a customer result

A scoped deployment example

Illustration: test a rear bay and a goods entrance with legitimate collections alongside staged restricted entry. Compare evidence usefulness with the supervisor's manual event log.

Acceptance-test approach

Agree event definitions and an independent log. Test both quiet periods and dispatch peaks, recording false alerts, missed events, delays and recovery after a network interruption.

Use the evaluation worksheet →

Available evidence

PGAK has published recorded demonstrations of its own (listed on the evidence page), each with its conditions and limits. No customer result or measured accuracy report is published yet, and the deployment scenarios are illustrations, not verified deployments.

PGAK’s own recording for part of this use case is shown further down this page, with its conditions and limits. It is a demonstration of one scene, not a measured result for your site; ask for evidence of the exact workflow you need.

Inspect evidence status →

Cost drivers and responsibilities

Camera count is only one input. Include processing, site preparation, configuration, integration, support, taxes and contract terms. Agree who maintains the feeds, reviews events and responds during an outage.

Understand project pricing →

Privacy and safety boundaries

Restrict shared footage to authorised security and operational purposes. Agree retention, visitor notice and controlled export; avoid unnecessary identification.

From a working install

Counting what crosses the line at the dock

A frame from a PGAK camera at a loading bay, taken during an evening unload. The vertical line is the counting boundary an operator drew, and the counters beside it move as things cross it.

The count climbing during an unload16s · no soundSixteen seconds of an evening unload. Each sack carried across the line is detected, given a tracking number so the same one is not counted twice, and added to the running total — the count goes from four to seven while you watch. The figure is there at the end of the unload instead of being reconstructed from memory, so it can be checked against the challan while the lorry is still at the bay.Conditions: Recorded 19 December 2022, 19:43, on an existing dock camera with one counting line. Confidence for each detection is shown on its box — the ones here sit around 0.84 to 0.90. Counting depends on the line being crossed cleanly: two people through the gap shoulder to shoulder, or a load carried back out and in again, are the cases to test on your own bay before relying on the figure.The transport company's name and telephone numbers, painted along the side of the lorry, are blurred for the whole clip. They belong to a third party and are not ours to publish. The clip carries no audio track.

The model is trained on a sack, and it counts sacks crossing a line. It is not an inventory system: it does not read a label, does not know whose consignment it is and does not reconcile against a stock ledger. What it gives you is a machine count of a known item type moving in and out at that bay — a figure to check a document against, not a replacement for counting the stock itself.

What a project here involves

Compatible cameras and recorders you already own are reused wherever they are suitable. Detection runs on an on-site processing unit, and camera suitability, stream access, processing hardware and network are all confirmed at the assessment — with anything extra itemised in the quote before you commit.

What the quote is made of

ComponentWhat decides itPrice
Cameras reusedWhich existing cameras give a usable view for each use case — confirmed on a still image of each view.Quoted per site
Cameras moved or addedOnly where a use case needs a view no camera has (plates and faces usually do). Each one named in the quote.Quoted per site
On-site processing unitDetection runs here, next to the recorder. Model, quantity and warranty depend on camera count and analytics.Quoted per site
Software / licencesWhich analytics on which cameras, the term, and the basis for renewal.Quoted per site
Installation and configurationStream setup, zones, schedules, alert routing and commissioning.Quoted per site
PilotAgreed scenes, acceptance criteria and duration before rollout.Quoted per site
SupportHours, response versus fix time, and who replaces failed hardware.Quoted per site
TaxesStated separately on the quote.Quoted per site

How it is evaluated

  1. 1. Self-check — Two minutes, nothing sent — find likely blockers.
  2. 2. Scope on paper — Cameras, use cases and acceptance tests, written down.
  3. 3. Site assessment — PGAK checks streams, views and network on your cameras.
  4. 4. Pilot against criteria — Precision, recall, false alerts and delay, agreed in advance.
  5. 5. Written proposal — Itemised as above; you decide whether to roll out.

Evidence for this page

  • Sacks counted across a line at a loading bay. Recorded 19 December 2022 on an existing dock camera with a single counting line. The transport company's name and telephone numbers are blurred throughout; they belong to a third party. Not proved: It counts a known item type across a line. It is not an inventory system and does not reconcile against a ledger. See it →

Discuss this use case

Send a callback request with this page’s context. Camera count is optional; do not include credentials or footage.

Request a camera assessment. Timing and scope agreed on the call. By submitting, you ask PGAK to contact you about this enquiry. Privacy notice.

Continue your evaluation