One of the most common questions we get from contractors evaluating FOBECON is some version of: "We have drawing sets from 2014 that were scanned from paper originals. Are those going to work?" The honest answer is: it depends on specific measurable properties of the scan, not on age or file format alone.
This article breaks down what we have learned from processing a range of scanned and digitally-originated PDFs. We want to be direct about what the system handles well, where it degrades, and what to do when you have drawings that fall into the problematic category.
What the System Actually Processes
When FOBECON receives a PDF, the first step is to determine whether each page is a native vector drawing or a rasterized scan. A native PDF from CAD software contains geometric objects: lines, arcs, polygons, and text elements stored as mathematical descriptions. A scanned PDF contains pixels. The processing paths for these two types are substantially different.
For vector PDFs, element detection operates directly on the geometry. Line endpoints, wall boundaries, opening positions, and annotation text are already structured objects. Accuracy for well-structured vector drawings is high, and processing is fast.
For scanned PDFs, the system runs image pre-processing before any element detection. This is where scan quality becomes determinative.
The Three Properties That Matter Most
Resolution
The minimum viable resolution for reliable automated element detection is 200 DPI. Below that threshold, fine-detail elements such as door swing arcs, thin partition walls, and small fixture symbols start to merge with each other or disappear into the background. The practical floor for adequate processing in most element categories is around 150 DPI; below that, automated takeoff is unreliable and the output requires manual correction that offsets the time saving.
Most modern office scanners default to 300 DPI and produce drawings that process cleanly. The problem case is drawings that were batch-scanned for archival purposes at 100 to 150 DPI, which was common in the early 2000s. If your archived drawing sets come from that era and those scanning settings, you will need to re-scan before automated processing will produce reliable output.
A practical rule: if you can comfortably read all dimension annotations and room labels on screen without zooming in more than 2x on an A1 sheet, the resolution is almost certainly adequate.
Skew and Geometric Distortion
When a drawing is physically placed on a scanner bed at a slight angle, the resulting image is skewed. FOBECON applies deskewing pre-processing automatically, and most skew within 2 to 3 degrees resolves without any manual intervention.
The harder problem is barrel distortion from flatbed scanners scanning a bound drawing set without fully flattening it, or from drawings photographed rather than scanned. Barrel distortion curves lines that should be straight, and the resulting geometric inaccuracy cascades into measurement errors. A line that curves by 4mm across 20 metres on an A1 plan represents a real-world measurement error of around 0.4 metres at 1:50 scale, which is consequential for wall linear metre counts and slab area calculations.
We can correct moderate barrel distortion algorithmically, but severe distortion from a cylindrical drawing roll photographed at an angle is beyond reliable correction. If you have drawing sets in that condition, re-scanning on a flatbed is the only reliable fix before processing.
Annotation Density and Overlay Legibility
The third quality factor is how clearly the drawing geometry separates from its annotation layer. In a well-structured drawing, dimensions, room labels, and element tags sit in defined zones that do not obscure the geometry they describe. In drawings with dense annotation, dimension strings sometimes overlap wall lines, grid labels print on top of structural columns, and small-scale details compress into illegible clusters.
Automated element detection operates on the same pixel field as annotation text. When text and geometry overlap significantly, the system's line detection can be confused by character shapes, and vice versa. The result is element counts with increased false-positive and missed-element rates in the dense-annotation zones.
This is not exclusively a scan-quality issue. It can also occur in vector PDFs with poorly structured annotation layers. But scanned drawings amplify it because the image pre-processing cannot easily separate annotation from geometry the way a structured vector file can.
What We Do With Degraded Input
When FOBECON processes a drawing page that falls below quality thresholds, the output includes a confidence score per element category and a quality flag on the affected pages. Estimating teams can use the confidence output to identify which pages need manual review rather than reviewing the entire output set.
For a mixed drawing set, for example a set where floors 1 to 5 are from a 2024 digital revision and floors 6 to 14 are from 2010 paper originals scanned at 150 DPI, the quality flags allow the team to accept the digital portions automatically and focus manual verification effort on the scanned portion. That is a significantly better outcome than re-doing the entire set manually, even though the scanned portions produce degraded output.
We are not claiming the system produces reliable takeoffs from poor input. The honest position is that input quality determines a ceiling on output reliability, and our confidence scoring is the mechanism that makes the ceiling visible rather than hidden.
Practical Assessment Before You Start
Before submitting a drawing set for automated processing, a quick manual check covers the main risk factors. Open a representative floor plan page and evaluate four things: Is all annotation text legible at 1x zoom? Are thin partition walls visually distinct from dimension extension lines? Do door symbols show clear separation between the door swing arc and the adjacent wall? Are grid labels sitting in the margin rather than overlapping gridlines?
If all four pass on a representative page, the drawing set is almost certainly in the range where automated processing will produce reliable output. If any fail, it is worth flagging the affected drawing types before processing so you know which parts of the output to scrutinise.
The more important point is that drawing quality assessment is a one-time exercise per project set. Once a team has processed a client's drawing standards once and understands how their files behave, subsequent projects from the same source can skip the assessment entirely.
A Note on Older Drawing Sets
Age alone is not the limiting factor. We have processed 1980s AutoCAD drawings exported to PDF that produced clean element detection, because the export preserved the vector geometry. We have also processed 2022 drawings that were scanned from paper proofs at 100 DPI because someone lost the digital originals. The scan process, not the drawing date, determines what the system receives.
For contractors with deep archives of older projects that they want to take into the automated workflow, the most reliable path is to identify the drawing sets by their scan properties rather than by date, and re-scan any that fall below the resolution and distortion thresholds before processing. The cost of re-scanning an A1 drawing set is low relative to the manual counting cost it replaces.
If re-scanning is not possible for a specific set, the confidence scoring in the output is the correct tool for managing the uncertainty. It does not make the problem disappear, but it makes the problem locatable.