Navigating Philippine Streets: Implementing YOLOv8 with Ghost Convolutions for Traffic Sign Detection

• Main Authors: Aquino, Alec Angelo A., Abu, Patricia Angela R, Alampay, Raphael
• Format: text
• Published: Archīum Ateneo 2024
• Subjects: Accuracy; Costs; Navigation; Computational modeling; Size measurement; Road safety; Real-time systems; Computer Engineering; Computer Sciences; Electrical and Computer Engineering; Engineering; Physical Sciences and Mathematics
• Online Access: https://archium.ateneo.edu/discs-faculty-pubs/426; https://doi.org/10.1109/ICCE-Taiwan62264.2024.10674537
• Institution: Ateneo De Manila University

This study implements YOLOv8 as a Traffic Sign Detection and Recognition System (TSDR) designed specifically for traffic signs indigenous to the Philippines. Dashcam footage containing traffic signs local to the Philippines were annotated to create a custom dataset comprised of 7 different traffic signs, and totaling to 1282 instances. It is imperative for a TSDR to function within environments characterized by limited computational resources. However, TSDR implementations often involve models that are resource intensive. There is a need to improve the efficiency of these models, specifically their inference speed for real-time detection. This study will present the integration of Ghost Convolutions into YOLOv8 in order to address computational scarcity and enhance performance speeds. Our experimental results yielded models that performed up to 40.09% faster, albeit with a marginal trade-off in accuracy.