Zhai Weiliang
ABSTRACT
To enhance driving safety and protect both motorists and pedestrians, this study developed a collision warning system for road-pedestrian detection using deep learning. The system employs the YOLO model as its core technology and improves pedestrian recognition through several enhancements, including data augmentation, model architecture optimization, and refined feature-extraction techniques. It calculates the distance between vehicles and pedestrians and, based on Time-to-Collision (TTC), establishes a multi-level collision-risk model. Corresponding warning strategies are then provided for each risk level to support timely driver response. The system was evaluated in both simulated and real-world road environments. Findings show that it achieves high accuracy and real-time performance in pedestrian detection and effectively issues early warnings when collision risks arise. A user-acceptability survey was also conducted to assess usability, compatibility, security, and overall system performance. Results confirm that the system offers valuable decision-support for drivers, contributes to reducing potential road-traffic accidents, and promotes overall road safety. Future work may focus on integrating the system with autonomous-driving technologies and optimizing its performance across diverse road conditions to further strengthen road-safety assurance.
Keywords: Distance measurement, Early warning system, Deep learning, Pedestrian recognition
https://doi.org/10.57180/cqgr8570