AI symbol detection for counting on dense plan sheets

Counting 180 sprinkler heads across four irrigation sheets is exactly the work that human attention is worst at. The symbols are identical, the density is high, and there is nothing to reason about — which makes it the strongest case for automated detection in takeoff.

How the workflow runs

1. Select an example symbol on the sheet — one inlet, one pole, one head.
2. Detection scans the sheet and proposes every match it finds.
3. Candidates appear as markers, each with a confidence indication.
4. You confirm the obvious ones in bulk and inspect the uncertain ones.
5. Rejected candidates disappear; accepted ones become counted items in the condition.

Where it works well and where it struggles

Run detection once per symbol type, not once per sheet. Mixing a curb inlet and an area inlet into one detection pass produces a total that has to be manually split afterwards — which loses the time you saved.

Reviewing by exception

• Look first at clusters — overlapping markers usually mean a double detection.
• Then at the sheet edges and match lines, where partial symbols confuse matching.
• Then at low-confidence candidates as a group.
• Finally, do one visual sweep looking for symbols with no marker.

Attributes still come from you

Detection tells you how many. It does not tell you that structures 4 through 9 are 12' deep, that half the poles are 25' and half are 30', or that the trees split three ways by species. Those come from labels, schedules and the profile — and they are what makes the count priceable.

The realistic gain is not perfect counts with no effort. It is turning a two-hour manual sweep into a fifteen-minute review pass, with attention spent on the genuinely ambiguous items instead of on the 170 that were never in doubt.