113. Enhancing Traffic Safety through Predicting Child Pedestrian Traj…

International Oral Session, International, 27(금) 13:20-15:00, 라카이볼룸Ⅰ 

Enhancing Traffic Safety through Predicting Child Pedestrian Trajectories: A Study Using Transformer-Based Deep Learning Approach / Sungmin You(Ajou University), Haneul Park(Ajou University), Chiwoo Roh(Ajou University), Sungeun Cho(Ajou University), Jaehyun(Jason) So(Ajou University)

Enhancing child pedestrian safety is crucial, as traditional post-accident measures are constrained by maintenance costs and infrastructure limitations. To address this, a proactive system using deep learning has been introduced. While Transformer-based models effectively predict pedestrian movements, research on child pedestrians is limited. This study employs the TUTR (Trajectory Unified Transformer) model to predict child pedestrian trajectories using AI-HUB video data, capturing various risky behaviors. The TUTR model showed effective performance, as measured by Average Displacement Error (ADE) and Final Displacement Error (FDE). Integrating this system into autonomous vehicles and traffic infrastructure could reduce accidents involving child pedestrians.