Real-Time Pedestrian Detection and Passive Signal Monitoring Using YOLOv12
Abstract
Pedestrian crossings are vital for urban safety and traffic management. However, fixed-time signal controls ignore real-time pedestrian demand, leading to inefficient traffic flow and underutilized infrastructure. To address this, this project introduces a Real-Time Pedestrian Detection and Passive Signal Monitoring Using YOLOv12. The core objectives were to develop a pedestrian detection and counting system, implement analytics functionalities, and evaluate overall performance using recorded footage. The proposed system was validated using real-world video and publicly available datasets. Experimental results demonstrated that the system successfully performs real-time pedestrian counting, demand classification, and adaptive traffic light control. Notably, the adaptive control strategy optimized traffic flow by dynamically reducing the pedestrian green-light duration from 8 seconds to 4 seconds when no active crossing demand was detected. This intelligent adjustment resulted in a cumulative time savings of approximately 2.15 minutes across 26 crossing cycles. The successful development of this system highlights the immense capability of integrating computer vision and deep learning for pedestrian monitoring and analytics applications.