September 2, 2026
Counting symbols on a floor plan is repetitive work with an unforgiving output. There are two hundred socket symbols spread across a sheet that also carries furniture, dimensions, hatching and text, they all look alike, and the number you write down has to be right.
Counts behave differently from measurements. A wall measured 2% short is usually a 2% error in that line. A missed light fitting is a whole fitting — the luminaire, its fixings, the labour to install it, and whatever the bill attaches to that point. This article is about counting symbols on a PDF drawing automatically rather than by eye: what a detection tool can genuinely tell you, and — the part that decides whether the count is usable — what it cannot.
It is not attention. It is repetition.
The sheet works against you. Symbols sit on top of hatching, cross dimension lines, and overlap the architectural background. Where the background is drawn at the same line weight as the services, a socket against a wall hatch is genuinely hard to separate from it — and a scan makes it worse.
Variations matter. A single socket, a double socket and a floor box can share a family resemblance and belong on three different lines of the bill. So can a standard downlight and an emergency one, differing by a small mark inside the same circle.
You lose your place. The usual defences — highlighting each one, working zone by zone, tallying in fives — exist precisely because the eye cannot hold that state on its own. They are also a large part of the hidden cost of manual takeoff.
Revisions bring the job back. Revision clouds are supposed to tell you what moved, and on a well-issued drawing they do. Where the cloud is missing, or the change is a symbol type rather than a position, they do not — and the safe response is to count the reissued sheet from scratch, as its own count, rather than trying to work out the difference.
Object detection reverses the work. Instead of finding every instance yourself, you identify one and let the tool find matches for it.
In Planmetry you select the object detection tool, click one example of the symbol, and the tool searches that page for symbols that look like it. It reports how many it found, you give the object a name and pick a colour, and the matches are added to the material list as counted items, marked on the drawing in that colour.
Two things about that are worth drawing out.
It works from what is drawn. Detection reads the appearance of the symbol rather than looking anything up in a fixture library, so nothing about it is tied to one trade's symbols. What it finds on a given sheet depends on how cleanly that sheet is drawn.
Pick a distinctive part of the symbol, not a wide area around it. A selection that takes in the wall behind the symbol, or a neighbouring dimension, is describing that context as part of what to look for. The clearest single feature gives the best result.
Wrong matches are removed from the item list one at a time, so a false positive is a correction rather than a restart. Detection runs on the page you are working on: a document holds one drawing page, so a set is counted sheet by sheet.
This is the part that decides whether you can trust the count.
Coloured matches tell you what was found. They tell you nothing about what was missed.
Forty-seven marked sockets on the screen looks like completeness. It is not: it is forty-seven confirmations, and no information at all about the sockets that were not marked. A symbol under a dimension line, rotated, partly covered by text, or drawn slightly differently in one room because it came from another drawing, can be missed — and nothing on the screen indicates its absence. If detection returns nothing, the tool says so, and that is a prompt to try a cleaner example rather than a conclusion that the sheet is empty.
Completeness is a separate question from anything on the screen, and these are the checks that answer it:
Detection takes out most of the repetition. It does not take out the review, and the time it frees is what the review is for.
Variants are separate quantities. Single, double and switched sockets are three items even where the symbols differ by a tick. On an office floor that is three detections and three rows — 38 twin, 12 single, 4 floor boxes — not one row of 54. You cannot put a rate against 54.
The same symbol can mean different things in different places. A downlight in a corridor and the same downlight in a wet room may be different products at different prices. The type reference and the specification settle that, not the symbol.
Not everything that repeats is counted. Conduit, ductwork, skirting and cable tray repeat visually and are priced by the metre. Their fittings, supports and terminations may be counted, derived from the run, or deemed included, depending on how the bill is written. Deciding what is counted, what is measured and what is neither is a decision taken before you start.
The material list holds counted and measured items together for the sheet you are on, and exports to Excel. What comes out is a quantity list: rates, labour and the assumptions behind them are applied where you price.
If you also need the runs — conduit, pipework, duct — measure them on the same sheet while you are in it. The PDF quantity takeoff workflow covers how the measured and counted parts of a package fit together.
Scans carry noise that makes similar symbols less similar to each other. Heavily overlaid sheets hide symbols in a background drawn as prominently as the services. Sheets without a legend leave the meaning of a symbol to inference, and a confident count of the wrong item is worse than an uncertain one. A drawing issued as preliminary can be counted, but the count describes a design that is still moving.
None of these is a reason to stop. On a live tender it usually means an RFI, or a stated assumption in the return — going direct to the designer is often not open to you. Either way the uncertainty is written down instead of absorbed into a number.
The useful test is a drawing you already know the answer for — one you have counted by hand. Count it again this way and compare. If the numbers disagree, the reason is the useful part: a legend symbol counted as an installed one, a variant grouped with its neighbours, a symbol detection missed or wrongly matched, or a genuine miss in the hand count.
