Drone roof inventory: the exact number of tiles

Counting tiles or measuring a roof means climbing onto it, with the risk and time that entails, or estimating by eye and burying the error in the quote. A twenty-minute flight solves both.

«A twenty-minute flight and a report with the exact number of tiles. No scaffolding, nobody on the roof, no margin of error in the estimate.»

Client

Contractors, building engineers, surveyors, heritage bodies, insurers

Input

Drone flight, 10 – 20 m above the pitch

Gimbal

Perpendicular to the roof plane

Resolution

≈ 0.5 cm per pixel (GSD)

Technique

100 % bespoke model, tiled inference

Output

Count, area, density per m², annotated PDF report

What you get

A piece count, the pitch area, density per square metre and a PDF report with the annotated image, piece by piece. It is the document that holds up a materials quote without a safety cushion.

Between two flights separated in time you can see what has shifted or broken. That turns a one-off measurement into preventive maintenance, and into evidence usable with an insurer.

Why a bespoke model is required

The Mallorcan roof tile is in no public dataset. None. The model is trained from scratch on images of real roofs, and that is exactly what makes the count reliable.

Flying between ten and twenty metres above the pitch, gimbal perpendicular to the plane, yields about 0.5 cm per pixel. That figure — the GSD, the centimetres of reality each pixel covers — is what makes counting pieces possible, and it is what justifies the flight altitude.

In scenes this dense, inference is tiled: the image is sliced, detection runs on each tile and the results are merged. Without that, a conventional detector saturates and stops counting.

Can be extended with

• Detection of broken or displaced tiles
• Flight-to-flight comparison for preventive maintenance
• Expert report for insurance
• Orthomosaic when one photo is not enough