Research Article
AI-Based Signal Control for Balancing Pedestrian Two-Stage Crossing Safety and Traffic Flow Efficiency
@INPROCEEDINGS{10.1007/978-3-032-38705-9_9, author={Wei-Hsun Lee and Po-Hao Huang and Che-Yu Chang}, title={AI-Based Signal Control for Balancing Pedestrian Two-Stage Crossing Safety and Traffic Flow Efficiency}, proceedings={Intelligent Transport Systems. 9th EAI International Conference, INTSYS 2025, Marietta, GA, USA, December 10, 2025, Proceedings}, proceedings_a={INTSYS}, year={2026}, month={10}, keywords={Human-centered Traffic Environment Two-stage Pedestrian Crossing Vulnerable Road User Protection Deep Reinforcement Learning}, doi={10.1007/978-3-032-38705-9_9} }- Wei-Hsun Lee
Po-Hao Huang
Che-Yu Chang
Year: 2026
AI-Based Signal Control for Balancing Pedestrian Two-Stage Crossing Safety and Traffic Flow Efficiency
INTSYS
Springer
DOI: 10.1007/978-3-032-38705-9_9
Abstract
Pedestrian safety at intersections has emerged as a pressing issue, particularly in the context of two-stage crossings that are intended to mitigate conflicts but often result in increased pedestrian delays and reduced traffic efficiency. This study develops an AI-based signal control framework designed to optimize the trade-off between safety and operational performance. The proposed method incorporates surrogate safety indicators, including Post-Encroachment Time (PET) and Time-to-Collision (TTC), into a learning-based optimization process that continuously adapts phase allocation to dynamic traffic conditions. Simulation experiments with real-world traffic data reveal that the system reduces pedestrian conflict events—measured by PET and TTC thresholds—by more than 30%, while maintaining vehicle travel time within 5% of the fixed-time baseline and decreasing average queue length by approximately 12%. Compared with conventional fixed-time and rule-based adaptive control strategies, the AI-based framework demonstrates superior capability in achieving a balanced improvement in both safety and efficiency. These findings show the potential of AI-enhanced signal control to advance people-oriented intersection management and contribute to sustainable urban mobility.