Main work and achievements:
WP4 made significant progress during the past six months, advancing AI-powered tools for improving road safety, monitoring driver behaviour, and detecting infrastructure maintenance needs. During this period focus was placed on further developing the real-time edge computing system for visual monitoring, with significant improvements in object detection accuracy for key safety indicators such as helmets, seatbelts, mobile phone usage, and license plates. The models were trained and refined using new datasets from multiple pilot sites, ensuring they can work reliably across diverse conditions. Alongside visual monitoring, work progressed on behavioural analysis to anticipate dangerous situations by identifying abnormal vehicle patterns and traffic violations, with initial experiments successfully capturing sensor data and detecting manoeuvres such as abrupt braking and zigzag driving.
Road infrastructure monitoring also advanced, with deep-learning models improved to detect potholes and other surface defects more accurately, supported by real-world field tests and the development of metrics to quantify defect severity. Environmental data integration progressed through weather modelling setup and evaluation of existing weather station networks to assess their relevance for road safety applications. Finally, AI-driven 3D scene reconstruction work began, using UAV-collected data and cutting-edge neural rendering techniques to generate detailed 3D models of roads and accident scenes, improving situational awareness.
Next steps:
WP4 will focus on expanding datasets, optimising detection and behavioural models, integrating environmental and visual data streams, and refining real-time monitoring capabilities. Field testing and pilot integration will be a key priority, including optimising camera placement, testing connectivity for embedded devices, and ensuring reliable system performance under challenging conditions. These efforts will strengthen WP4’s role in enabling predictive road maintenance, enhanced safety monitoring, and real-time decision-making across all pilot sites.
