What AI can see on a construction site
Construction sites generate thousands of photographs every week. Site engineers take photos of completed work for records. Clerks of works photograph defects. Project managers capture progress. Safety officers document compliance and non-compliance. Most of these photographs are stored in folders, attached to emails, or uploaded to project management platforms where they are rarely reviewed systematically.
AI changes this. Computer vision models can process site photographs and extract structured observations — detecting defects, assessing progress, identifying safety concerns, and categorising the condition of completed work. The photographs you are already taking become a data source.
But let us be precise about what current AI can and cannot do with construction site photographs.
What AI can detect reliably:
- Visible cracks in concrete, plasterwork, and masonry (surface cracks wider than approximately 1mm)
- Water damage: staining, pooling, damp patches on surfaces
- Missing or incomplete elements: gaps in insulation, unfixed ceiling tiles, absent fire stopping
- General alignment issues: visibly out-of-plumb walls, uneven floor surfaces, misaligned ceiling grids
- PPE presence: hard hats, high-visibility vests, safety footwear (from medium-range photographs)
- General site condition: housekeeping, material storage, access routes
What AI cannot reliably detect:
- Hairline cracks (below approximately 0.5mm) — they do not resolve in standard site photographs
- Structural adequacy — a wall can look perfectly fine in a photograph and be structurally inadequate
- Hidden defects — anything concealed behind finishes, in cavities, or below grade
- Specification compliance — whether the installed material matches the specification (AI can see what it looks like, not what it is made of)
- Workmanship quality to tolerance standards — whether a surface is within the specified flatness tolerance requires measurement, not visual inspection
