AI for road safety in LMICs
for a brighter
tomorrow
Responsible AI
Research at a glance
My research sits at the intersection of computer vision, machine learning, and intelligent transport systems. I develop AI methods that turn road images and video into useful safety information, with a particular focus on real-world conditions where data can be noisy, imbalanced, incomplete, or difficult to label. My work includes multi-task visual classification of road infrastructure and traffic attributes, transformer-based vision models, multimodal and vision-language methods, and predictive modelling of road safety risk. I also work with satellite imagery and geospatial data to extend analysis beyond street-level cameras. A key part of my research is making models more robust, scalable, and useful for practical deployment rather than only improving benchmark accuracy. The broader goal is to build AI systems that can understand complex physical environments and support safer, data-driven decision making.
Research objectives
- Develop robust computer vision models for automatic road and traffic attribute recognition from images and video.
- Improve multi-task learning under class imbalance, noisy labels, and limited real-world data.
- Combine vision transformers, multimodal models, and vision-language models for richer scene understanding.
- Build predictive models linking visual road features to crash risk and road safety outcomes.
- Extend analysis using satellite imagery and geospatial data for scalable transport and infrastructure monitoring.
Publications
| Title | Authors | Venue | Year | Link |
|---|---|---|---|---|
| Auditing iRAP's ViDA Risk Engine: A Two-Stage Surrogate Learning and Orthogonalized Heterogeneity Framework for Modelled Road Safety | Hassani, A. First author | Infrastructures | 2026 | ↗ |
| Multi-Scale Spatio-Temporal Feature Aggregation for Road Safety Attributes Classification From Videos | Hassani, A. Co-author, with University of Zagreb and iRAP | IEEE Access | 2026 | – |
| Optimizing Car Collision Detection Using Large Dashcam-Based Datasets | Hassani, A. Co-author | Applied Sciences | 2025 | – |
| A Spatio-Temporal Multi-Task Framework for iRAP Road Attribute Classification from Street-Level Videos | Hassani, A. Co-author | ACAI 2025 (IEEE) | 2025 | – |
Conference contributions
| Title | Conference | Date | Link |
|---|---|---|---|
| EfficientNet-Swin Transformer for Automated iRAP Road Safety Attribute Extraction in Low- and Middle-Income Countries | ICTR 2025 | 2025 | – |
| Compositional Diffusion Image Synthesis for Long-Tailed Multi-Country Road Infrastructure Assessment | 8th IRTAD International Conference | 2026 | – |
| Quantifying Signal-Gating Error in Queue Delay Models Using Video-Derived Trajectories | EAI FABULOUS 2026 | 2026 | – |
Secondments & collaborations
| Host organisation | Period | Purpose | Status |
|---|---|---|---|
| Verne (Project 3 Mobility – P3M) Zagreb, Croatia | – | Autonomous mobility and robotaxi technology company | – |