ADVANCED PEDESTRIAN DETECTION AND INTELLIGENT TRAFFIC MANAGEMENT VIA DEEP LEARNING

Authors : Rehnüma Küçükilhan Turunç, Ahmet Haşim Yurttakal

Pages : 707-720

Doi:10.46519/ij3dptdi.1782019

View : 81 | Download : 224

Publication Date : 2025-12-28

Article Type : Research Paper

Abstract :In this study, we developed a deep learning-based pedestrian detection system to prevent pedestrian collisions. These collisions account for a significant portion of urban traffic accidents. We collected and annotated a custom dataset of 620 high-resolution pedestrian images using the MakeSense labeling tool. Using this dataset, we trained YOLOv8, YOLOv11, and YOLOv12 models and evaluated them based on precision, recall, mAP, and F1-score. The training processes were conducted in the Google Colab environment using Python, supported by GPU acceleration. Among the models, YOLOv11-S achieved the highest performance with an F1-score of 94.9%. We then integrated the trained model into a PyQt5-based desktop simulation interface, enabling real-time pedestrian detection and automated traffic light control. The results demonstrate that deep learning-based pedestrian detection systems can operate effectively in real-time scenarios and provide a sustainable, scalable solution for smart city infrastructures.
Keywords : Computer Vision, Deep Learning, YOLO, Traffic Management, Pedestrian Detection

ORIGINAL ARTICLE URL