Expert article
Artificial intelligence in construction: state of the art, not a buzzword
A great deal is promised about AI on the construction site. An honest look separates three things that run today from two that come out of research.
Expert article by Open Experience GmbH. As of: September 2026
Where is AI used on the construction site today?
AI is in productive use wherever large quantities of images have to be evaluated automatically: in detecting people and faces for pixelation that complies with data protection law, in detecting relevant objects that are to remain exempt from pixelation, and in classifying images automatically by building elements such as doors or windows. Further advanced, but still out of research, are the automatic detection of visual defects and position finding on the construction site.
Legal framework: Germany (Civil Code BGB, construction contract rules VOB/B, fee schedule HOAI). Other countries have different rules, and contractual agreements take precedence over the standard periods named here.
What runs — and what is research
The dividing line does not run between simple and complex, but between tasks with an unambiguous answer and tasks that call for judgement.
| Application | Status | What it delivers |
|---|---|---|
| *Person and face detection* | In production | Automatic pixelation before the image goes into storage |
| *Detection of relevant objects* | In production | Measuring instruments and tools stay exempt from pixelation |
| *Automatic image classification* | In production | Filtering by recognised categories instead of by file name |
| *Detection of visual defects* | Out of research | Marking scratches, cracks and damage in the image |
| *Localisation on the construction site* | Out of research | Determining the capture position automatically instead of by hand |
"Out of research" means: demonstrated and trialled in the ESKIMO project, not available as a series function.
The unspectacular gain: pixelation
The most valuable AI application in production on site is not one that people give talks about. Site photographs show people at work; without automatic pixelation, comprehensive photo documentation is hard to sustain under data protection law. Yet that is precisely what decides whether documentation is complete at all.
- Two modesEither whole people or only faces are made unrecognisable — depending on what the project requires.
- Exceptions are necessaryA laser distance meter in someone's hand is an item of evidence and must not be pixelated along with them. That is why the system recognises objects as well.
- At source, not afterwardsPixelation happens before storage. What is never stored in the first place does not have to be deleted later.
The ESKIMO project
ESKIMO stood for the development of artificial intelligence system components for a digital, mobile value chain in construction execution. Funded by the Federal Ministry of Education and Research, started on 1 April 2020, running for 24 months, total costs of 2.4 million euros, eleven partners from the construction industry and research — coordinated by Open Experience. Three strands of results are relevant to construction execution.
- Technical quality assuranceDetection of visual deviations such as damage, staining and discolouration, and of structural differences from the BIM model.
- Commercial quality assuranceReconciling the model with what has been built: regular as-built captures make it possible to determine when individual elements were installed.
- Intelligent construction logisticsOptimising storage areas and haul routes on and around the construction site.
Before that: digiBau — the hardware it takes
AI evaluation presupposes images that come into being in the first place. The digiBau project (started 1 July 2017, running for 24 months, funded by the Federal Ministry for Economic Affairs and Energy, together with Darmstadt University of Applied Sciences) produced a modular system for exactly that: a sensor and camera module on the safety helmet, a system unit for image and sensor evaluation, and a mobile "Smart Inspection Cockpit" for the person on site. Open Experience was responsible for the software components. Today's helmet attachment grew out of that work.
What AI in construction does not deliver
No causes
A crack that has been detected is a symptom. Why it is there is for a human to establish.
No responsibility
Who has to answer for a defect is a question of contract, not a question of images.
No assessment
Whether a deviation is acceptable is decided by the specialist designers, not by the model.
Nothing without images
Every evaluation presupposes regular, located captures. Where the capture is missing, no algorithm helps.
No acceptance
Acceptance is a legal act between people and remains one.
But
Anything that has to look at every image with equal thoroughness is a job for a machine.
What this means for practical use
Anyone who wants to use AI in construction starts not with the AI but with the capture: regular, located images of comparable quality. That is the precondition for every evaluation — and the part that takes discipline. In legal terms the General Data Protection Regulation governs its use, and since 2024 the EU AI Act as well; for the applications described here it is above all the data protection compliant processing of images of people that matters.
Frequently asked questions
Does the software detect defects automatically?
Not in production. The automatic detection of visual defects was demonstrated in the ESKIMO project and is being developed further. What works in production today is pixelation, object detection and image classification.
Is my project data used for training?
No. Training data comes from research holdings and from released material. Project data stays in the project — with construction documentation that is also the only workable route under data protection law.
Does AI need a special camera?
No, but it does need comparable images. Regularity, lighting and fixed capture points have a stronger effect than resolution.
What does the EU AI Act change for construction companies?
It classifies AI systems by risk and attaches obligations to that classification. For photo documentation with pixelation, transparency and data protection are the decisive points — this article is no substitute for legal advice.
Does AI replace the site manager?
No. It takes the uniform inspection work off their hands — looking at every image with the same thoroughness. Judging, deciding and negotiating stay with people.
How do you begin?
With the capture. Without regular, located images there is nothing to evaluate — with them the benefit arises before any AI comes into play.
Sources and legal basis
The legal statements in this article are based on the primary sources listed below. The article is not a substitute for legal advice in an individual case.
The capture first, then the automation.
45 minutes on your project: we show what happens automatically today — and what makes the difference before AI even comes into play.