At the 20th International Road Safety on Five Continents (RS5C) Conference, held in Leeds from 3–5 September 2025, colleagues from the National Technical University of Athens (NTUA) presented research on predicting pedestrian violations at urban intersections using artificial intelligence.

The study, which is part of the PHOEBE project, addressed the challenge of illegal crossings in Athens, a city with heavy pedestrian traffic but limited monitoring infrastructure. NTUA employed computer vision tools and machine learning to analyse pedestrian behaviour at two key intersections in central Athens.

Using smartphone cameras for data collection, the team applied a video detection framework combining YOLOv8, ResNet-50, and Kalman filtering to track movements relative to crosswalks and traffic lights. This data was then used to train two machine learning models: Random Forest and XGBoost.

Results showed that although XGBoost required more training time, its superior precision and balance across classes make it a more reliable tool for urban safety applications. NTUA emphasised the potential of such models to be integrated into real-time monitoring systems, early-warning tools, and policy evaluations. Future work will expand to include contextual factors such as weather, road layout, and pedestrian demographics, to scale the approach across cities and enhance traffic light compliance.

PHOEBE and NTUA invite you to download the presentation and the full text here.

After the first interview with our colleagues from the Athens pilot, we continue our series with several questions about the current progress of models and tests in the West Midlands, which are mostly focused on modelling the potential impact of speed reduction on the overall safety of vulnerable road users. Additional research updates, which were based on the West Midland pilots, were recently posted on our R&D blog.

What do your findings suggest are the main impacts to expect from decreasing the speed limit on all 40 mph roads?

Our results suggest that reducing the speed limit will lead to a 45% reduction in the number of people killed or seriously injured every year. No segments were awarded the lowest risk rating in this analysis.

Our analysis predicts a significant increase in the number of journeys carried out on foot, by around 50%, suggesting that more people will gain the positive health benefits that come from being active.

What did we learn about cycleways in the West Midlands?

The Bristol Road (A38) cycle way is a relatively new path for cyclists to take that runs alongside a main road. Local authorities in the West Midlands have many similar schemes in the pipeline, including an extension to the south of the Bristol road.

We chose to look at the area in terms of the situation as it is at the moment, and what would happen if the cycle lane were suddenly removed. By taking this approach, we could attempt to understand the impact of introducing similar cycle lanes in the West Midlands.

We also decided to perform a ‘CycleRAP assessment’ of the Bristol road, which involved a member of the team cycling the route with a handlebar-mounted smartphone that uses the Lane Patrol app to take photographs every 10 metres of travel. These photographs were used to create risk models of the road and cycleway. The subsequent analysis of these models suggests that the number of people who are killed or seriously injured along this route dropped to around a third of the level that existed before the cycleway was built.  

What feedback have you received from the local stakeholders?

We held a road safety workshop in collaboration with Transport for the West Midlands (TfWM), where the results of the active travel studies were presented and discussed. These discussions helped to spread awareness of new technologies and data products that can be utilised by transport planners working for local authorities.

During 2025, TfWM participated in a series of working meetings to review our results, which provides the project with the context and understanding that can only come from local experts. Most recently, TfWM have started the process of determining how insights from PHOEBE can be distributed to the relevant working groups.