Application of YOLOv5 for Automatic Parking Lot Detection and Monitoring System

Abstract

The rapid growth of vehicles in urban areas has led to parking space limitations and longer search times for available spots. This study aims to design and implement an automated parking availability detection and monitoring system using the YOLOv5 algorithm and image input from drone cameras. The methodology includes image data acquisition through drone recordings from two different viewpoints (top-down and side view), data labeling, object detection model training, and classification of parking slot status (vacant or occupied). System evaluation was conducted by measuring precision, recall, accuracy, and mAP@0.5. The testing results show that camera angle affects detection accuracy: from the side view, the system achieved 100% precision, 75.86% recall, and 75.86% mAP@0.5, while from the top-down view, recall and mAP@0.5 dropped to 35.29% and 35.00%, respectively. These findings are supported by the Confusion Matrix and Precision-Recall Curve visualizations. The developed system successfully detects and monitors parking slot availability in real-time, displaying the results via a digital dashboard. The use of drone cameras enables broader and more flexible area coverage compared to static cameras. Therefore, this system has the potential to be a practical solution in the development of deep learning-based smart parking in public areas.