In-line parcel dimensioning station with a depth camera

Maple Freight Hub charges e-commerce shippers by dimensional weight: length × width × height divided by a carrier divisor, or actual weight if that is higher. Shippers declare dimensions when they book. An audit of 2,000 parcels last month found that 23% were declared smaller than they really were, and the hub loses roughly CA$41,000 a month in under-billing at Brampton alone.

The operations manager wants an automatic dimensioning station on each induction line, before parcels are sorted. Commercial units are expensive, so the engineering team wants to understand the problem properly before choosing one, and possibly build a first unit in-house.

The proposed station: - an overhead depth camera 2.2 m above a belt running at 1.5 m/s; - a photo-eye that triggers a measurement as each parcel enters; - an in-line scale; - a message to the warehouse management system (WMS) with parcel ID, dimensions, weight and a confidence flag.

Parcels range from 100 × 100 × 20 mm envelopes to 1,200 × 800 × 800 mm boxes. Some are soft poly bags, and some arrive touching each other. The target is ±5 mm on boxes, which is close to what legal-for-trade dimensioners must meet in Canada and the US, and a clear "cannot measure" flag rather than a wrong number when the station is unsure.

You will: - set a measurement and timing budget; - generate synthetic depth frames of parcels with realistic noise, since you may not own a depth camera; - write point-cloud processing that removes the belt, separates parcels and fits a box; - measure accuracy and repeatability against ground truth; - write the station logic for trigger, scale and WMS message at 1,800 parcels an hour; - estimate recovered revenue for both hubs.

Done looks like a processing pipeline that meets ±5 mm on rigid boxes in your tests, flags what it cannot measure, keeps up with the line, and a report the operations manager can use to decide between building and buying.