Pedestrian-Based Adaptive Traffic Light Control Using YOLOv8
G. Ortega, Amiel Mir Odiamar Ordas, Jocelyn F. Villaverde, Roben A. Juanatas, Irish C. Juanatas · 2024
The advancement of YOLO and similar computer vision models represents a technological era that could revolutionize numerous aspects of day-to-day life, including traffic management. This study explores the feasibility of utilizing YOLOv8 for pedestrian detection and counting to control traffic light timing sequences at isolated pedestrian crossings. By employing the Raspberry Pi 4 Model B with a camera module, real-time pedestrian video data can be captured and processed through YOLOv8. For the hardware, three LEDs and a Dot Matrix Module replicate conventional traffic light signals and timing, effectively simulating a traffic light system. The capabilities of the custom YOLOv8 model are evaluated through pedestrian detection accuracy, calculated by percentage error across twenty samples. Results reveal an average percent error of 7.33%, indicating 92.67% accuracy in pedestrian detection and counting. These findings underscore the model's competence for pedestrian-based adaptive traffic light control, highlighting its potential to enhance practicality and safety at pedestrian crossings by dynamically adjusting signal timings based on real-time pedestrian activity.