AN AI AND V2X-BASED SMART INTERSECTION SYSTEM FOR VULNERABLE ROAD USERS WITH REDUCED MOBILITY OR PERCEPTION
IX. BASKENT INTERNATIONAL CONFERENCE ON MULTIDISCIPLINARY STUDIES , Ankara, Türkiye, 11 - 12 Eylül 2026, ss.191-211, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Basıldığı Şehir: Ankara
- Basıldığı Ülke: Türkiye
- Sayfa Sayıları: ss.191-211
- Ankara Üniversitesi Adresli: Evet
Özet
This paper proposes an artificial intelligence (AI) and V2X communication-based smart intersection system aimed at improving crossing safety for vulnerable road users with reduced mobility or perception. In this study, four representative subclasses are addressed: wheelchair users, visually impaired pedestrians, elderly individuals, and mobility-restricted persons. These users require longer and more protected crossing time than conventional fixed-time signals provide. Existing roadside systems either detect such pedestrians without conveying the hazard to approaching vehicles, or convey signal phase and timing rather than the hazard event itself, leaving a gap between detection and driver awareness at urban intersections. To address this gap, the proposed system integrates a deep-learningbased (YOLOv8) detection module for the four subclasses, a genetic-algorithm-driven adaptive signal controller, and a V2X communication layer (V2I) that generates ETSI-compliant safety messages (DENM). By unifying perception, control, and communication under a single real-time framework, the system provides proactive hazard awareness before a pedestrian enters the vehicle's field of view. This is particularly critical for users who cannot visually detect an approaching vehicle, such as visually impaired pedestrians, or who cannot execute a last-moment avoidance manoeuvre, such as wheelchair users and mobility-restricted individuals. The detection module achieves a macro-average F1-score of 0.95 on a held-out test set, while pilot field evaluations at signalized intersections demonstrate a systemlevel end-to-end alert latency of 122 ms, an RSU-OBU packet delivery success rate of 97%, and a 28% reduction in crossing delays for the target pedestrians in the observed field scenarios. These results indicate that the proposed approach establishes an effective end-to-end cooperative safety chain and offers a scalable model for future intelligent transportation systems.