Every site survey produced hundreds of images, and an engineer traced each plot boundary by hand before it could be measured. It was slow, it varied from person to person, and it held up every report that depended on it.
Case study — Civil engineering
Automated delineation mapping from survey imagery.
How a survey team stopped tracing site boundaries by hand — and started checking a model’s work instead.
Client
Civil engineering consultancy (name withheld)
Disciplines
Computer vision, software
Engagement
Feasibility study, build, support
Delivered
Boundary-detection model, measurement pipeline, review tool
Fig. 1 — A typical survey site: rebar, spoil heaps and uneven light.
01The problem
02Our approach
We began with a fixed-price feasibility study on past surveys the team had already traced by hand, so the model was judged against their own work. Once its boundaries matched closely enough, we added a measurement step and a simple review screen so engineers could correct the few it got wrong.
“The engineers check the boundaries now. They don’t draw them.”
Survey lead, client team — quote to be confirmed
PLOT B · 412 m²
Site survey, auto-traced
6 boundaries · 0.4 m accuracy
CHECKED
Fig. 2 — A traced and measured plot, ready for an engineer to check.
03What we delivered
A segmentation model that traces plot boundaries from survey imagery, a pipeline that turns those boundaries into measured areas, and a review tool where engineers accept or adjust each result before it goes into a report.
04The result
Boundary tracing has moved from a manual step to a checked one. Replace this paragraph with the measured outcome once the client has approved the figures.