AI for road safety monitoring and crash prediction from micro- to macro levels
for a brighter
tomorrow
AI for Proactive Infrastructure Safety Management
Research at a glance
This research aims to move road safety analysis from reactive, crash-based approaches toward proactive, AI-driven monitoring. It investigates how in-vehicle telematics data can be aggregated onto road network nodes and edges for spatial analysis, and how sophisticated AI tools may be trained in a self-supervised way, without needing a target label, so that the network can learn purely from telematics and infrastructure data, without relying on crash data at all.
The work explores how the outputs of these AI tools can be used to partition the road network into areas with distinct patterns, and how a multiscale mechanism can be built into the model so that spatial relationships are learned by taking spatial proximity into account, allowing the framework to move beyond a single scale of analysis.
Research objectives
- Develop a robust methodology for aggregating in-vehicle telematics data onto road network nodes and edges.
- Train deep learning models in a self-supervised manner, without crash data or target labels, to generate spatial representations of the road network from telematics and infrastructure data alone.
- Use these AI-derived representations to partition the road network into areas with distinct driving-behavior patterns, extending the approach from nodes to edges.
- Incorporate a multiscale mechanism into the model so that spatial relationships are learned by accounting for spatial proximity, rather than assuming a single fixed scale.
- Validate the resulting partitions to assess how well this proactive, crash-data-free framework aligns with observed road safety outcomes.
Publications
| Title | Authors | Venue | Year | Link |
|---|---|---|---|---|
| Learning-based methods for spatial road safety analysis using in-vehicle telematics data: A systematic review | Paradiso, S., Ziakopoulos, A., Yannis, G. | Journal of Safety Research Vol. 97, pp. 737–761 | 2026 | ↗ |
Conference contributions
| Title | Conference | Date | Link |
|---|---|---|---|
| Combining diverse data sources for intersection crash analyses based on incomplete records | Navigating the Future of Traffic Management (IRF & ICCS) Athens, Greece | Jun–Jul 2025 | – |
| Aggregating Telematics for Road Safety Analysis | 12th International Congress on Transportation Research (ICTR) 2025 Thessaloniki, Greece | Oct 2025 | – |
| The use of Graph Neural Networks for Clustering in Road Safety Analysis | 13th Symposium of the European Association for Research in Transportation (hEART2025) Munich, Germany | Jun 2025 | – |
| Hierarchical Clustering on Graph Embeddings: A Scalable Approach to Risky Intersections | Road Safety on Five Continents (RS5C) Leeds, UK | Sep 2025 | – |
| Probabilistic Modeling for Node-Based Partitioning of Telematics-Informed Road Networks | 8th IRTAD International Conference Athens, Greece | 2026 | – |
| A Dual Graph Framework for Edge Embedding and Clustering in Road Safety Analysis | Transport Research Arena (TRA) 2026 Budapest, Hungary | May 2026 | – |
| A Graph Transformer Approach for Modeling Crash Occurrence at Intersections Using Telematics-Informed Road Networks | Road Safety and Simulation 2026 (RSS2026) Napoli, Italy | Jun 2026 | – |
| A Time-Window GNN-Based Network Partition for Identifying High- and Low-Risk Nodes in Road Networks | World Conference on Transport Research (WCTR2026) Toulouse, France | Jul 2026 | – |
Secondments & collaborations
| Host organisation | Period | Purpose | Status |
|---|---|---|---|
| Verne Zagreb, Croatia | Dec 2025 – Mar 2026 | – | Completed |