Thousands of kilometres of roads, only a handful of maintenance teams and no reliable automated way to find the worst damage. Through TEF Ostrava’s AI-MATTERS, the Czech AI-based road defect detection solution has undergone rigorous experimental testing. And the results challenge the status quo.
Imagine you are driving a van equipped with a conventional dashcam. As you drive, an artificial intelligence system silently analyses the image, recognising potholes, drainage channels and sewer grates in real time and tagging each one with GPS coordinates, a reliability score and a structured database entry. No manual inspection. No hours spent replaying records.
This system was developed by the Signal Lab at the Technical University of Ostrava and is currently being experimentally tested within the AI-MATTERS project at the Czech node in Ostrava, in close cooperation with FUTTEC, a company active in the field of road infrastructure maintenance.
A problem for which commercial tools are not enough
Road managers across Europe are facing the same frustration – existing commercial AI tools for automatic road surface inspection promise a lot but actually detect only the minimum. The question is not whether AI can detect defects on roads, but whether it can do so reliably enough to replace or meaningfully complement manual inspection.
To answer this question, the Ostrava team proposed direct comparative testing.
Five methods, thousands of hours of videos
A standardized test set of 50 road scenes with confirmed defects was used to compare the five approaches.
| Method | Detected faults | Detection success rate |
| Manual control (human reference) | 44 / 50 | 88 % |
| Commercial Solution 1 | 8 / 50 | 16 % |
| Commercial Solution 2 | 9 / 41* | 22 % |
| AI-MATTERS TEF candidate model 1 | 43 / 50 | 86 % |
| AI-MATTERS TEF candidate model 2 | 45 / 50 | 90 % |
The results were substantial. The two candidate models developed and tested in the AI-MATTERS project not only achieved comparable results to manual inspection, but the best model even outperformed it. In fact, it detected 90% of defects compared to 88% for trained human inspectors. In contrast, both commercial solutions detected less than a quarter of the samples.
“These figures confirm our assumption that universal commercial tools are not designed for the specific conditions of Central European road networks. Our models, trained on real data from Czech roads and using advanced architectures such as RT-DETR with hybrid matching during training, are tailor-made for this task,” says Dominik Vilímek from VŠB – Technical University Ostrava.
Damage measurement beyond detection
Pothole detection is only the first step. The real value for road managers is knowing how big the damage is, how fast it is increasing and when to schedule a repair.
Therefore, the second phase of testing in the AI-MATTERS project focuses on defect parameterization, which automatically calculates geometric characteristics such as length, width and area directly from the model output. The team is also experimenting with diffuse generative AI models to create richer defect representations, handle noise or incomplete data, and support downstream tasks such as predictive maintenance planning.
“We are moving from saying ‘there is a pothole’ to saying ‘there is a defect measuring 38 × 22 cm, which has increased by 15% since the last survey’. And that’s exactly the type of data that a maintenance planner really needs,” adds Vašek Mlynářík from FUTTEC.
Testing across hardware profiles
Industrial deployment means running on any available hardware, from cloud servers with GPUs to edge devices with just CPUs placed in service vehicles. The test program from AI-MATTERS TEF Ostrava systematically compares individual models on different hardware configurations and measures latency, throughput and resource utilization. The goal is to provide clear recommendations for deployment that match real-life operational constraints.
The third phase of testing verifies the complete integration layer (Docker containerization, API performance, stress testing and basic security checks) and ensures that the solution is not only accurate but also ready for production deployment.
What does this mean for European industry?
The AI-MATTERS testing infrastructure has enabled something that most SMEs cannot do on their own – rigorous, standardised and hardware-diverse comparison of AI models against commercial alternatives and human benchmarks. The results have been documented in validation reports in accordance with AI TEF Manufacturing standards, allowing for reusability and portability between different road networks and operational environments.
For FUTTEC and companies like it, this means access to AI that actually works – tested, validated and ready to deploy.