Takenaka

AI vision pallet verification at end of line

Automated counting of every pallet layer, a blocking release gate and full photographic traceability. Shipping errors are caught before parts ever leave the shop floor.

shipping errors after the release gate went live

100 %

traceability per pallet and operator

operator interface languages

Context

The problem

No independent verification layer

Every layer was counted by a single operator, with no second check and no confirmation before the pallet was released.

Zero traceability in a dispute

In the event of a customer claim or inventory discrepancy, the shop floor had no image, no timestamp and no operator identifier to fall back on.

Fragile counting sessions

An interruption mid-pallet (shift change, supervisor call) forced a restart, with a risk of double counting or a gap in the sequence.

Errors caught too late

Counting discrepancies only surfaced after the truck left the dock: as customer complaints or stock variances.

Solution

Deployed architecture

Capture

Camera tablet at the pallet station

Real-time WebRTC stream to the operator tablet. A photo of each layer, or a video mode that auto-advances once the count is confirmed.

Inference

YOLOv8 trained per product reference

A dedicated model per product family. Count and annotated image returned within seconds. Oriented bounding boxes supported for rotated parts.

Control

Mandatory release gate

Dual barcode scan of the pallet: no exit is possible without a match. No gate bypass on the application side.

Audit

Immutable log per action

Count, scan, operator login and tablet serial number all timestamped. Supervisor dashboard with status filters and per-pallet-code history.

Results

Measured results

Measured in production conditions on the deployed scope.

Shipping errors

After the release gate was rolled out across all pilot lines.

100 %

Pallet traceability

Every layer retained with image, operator and timestamp: audit-ready for compliance.

Supported languages

Japanese, English, Hindi, Nepali, Portuguese, Tagalog and Vietnamese: automatic switch at operator login.

Device-level

Accountability per terminal

Every session tied to the tablet serial number: faster internal-audit investigations.

Lessons learned

What we took away

The re-scan gate is the operational lever: vision alone is not enough without a physical block on pallet exit.

Session persistence across shifts avoids re-counts and discrepancies at handover.

A model per product reference is required as soon as geometries or stacking patterns differ significantly.

In-app packing guides cut back-and-forth trips to paper binders at the station.

Plan for 7 languages from the design stage if the shop floor is multicultural: the marginal cost is low once the i18n architecture is in place.

Get started

A similar case on your site?

Tell us about the process, the data available and your deployment constraints. We will assess feasibility within 48 hours.