Purpose: This study aims to develop and evaluate a low-cost vision-AI-based roadside unit (RSU) simulator integrated with SUMO to improve adaptive traffic signal control and support accessible pedestrian crossings for visually impaired individuals. Design/methods/approach: The proposed system integrates a Raspberry Pi 4 edge platform, camera-based vehicle detection using YOLOv5s, and SUMO traffic simulation through the TraCI interface. Its performance was evaluated against fixed time and SUMO adaptive control strategies under varying traffic conditions. The evaluation considered vehicle waiting time, queue length, travel time, pedestrian crossing success rate, and edge-computing performance. Findings: The proposed system consistently demonstrated better traffic efficiency than the baseline strategies, resulting in shorter vehicle waiting times, reduced queue lengths, and improved travel times in the evaluated scenarios. The accessibility-aware crossing mechanism also achieved a higher pedestrian crossing success, suggesting its potential to accommodate visually impaired pedestrians in adaptive traffic management. Furthermore, the edge-based perception module demonstrated promising detection performance and processing responsiveness, indicating its suitability for further investigation in resource-constrained environments. Research implications/limitations: The findings highlight the potential for integrating traffic efficiency and pedestrian accessibility within intelligent transportation systems. However, the evaluation was limited to a simulated single-intersection environment, requiring further real-world validation to establish operational reliability and accessibility of the proposed method. Originality/value: This study contributes an integrated simulation framework combining edge-based Vision-AI traffic perception, adaptive signal control, and accessibility-aware pedestrian assistance, thereby extending conventional vehicle-oriented traffic optimisation approaches toward more inclusive intersection management.
Copyrights © 2026