Smart Parking Space Availability Detection Using Aerial Images and YOLO-Based Deep Learning Models

Precise and timely parking space occupancy detection is an essential part of contemporary smart city infrastructure. In this paper, we introduce a strong deep learning-based car parking detection system with the YOLO (You Only Look Once) family of object detection models-specifically YOLOv5, YOLOv7, YOLOv8 and YOLOv9. Our aim is to label parking spaces as vacant or occupied using aerial-view image data with bounding box annotations in both COCO and YOLO formats. We preprocess the data by translating COCO-style annotations to YOLO format, then training both YOLO versions under identical experimental settings to ensure comparison consistency. The training process is streamlined for quicker execution with lower image resolution $(128 \times 128)$ and fewer epochs (20) while still having competitive detection accuracy. Comparative tests indicate that newer YOLO versions provide better mean average precision (mAP) and more accurate bounding box localization under various lighting and occlusion scenarios. The system being proposed can be implemented on GPU-accelerated edge devices, which results in it being an efficient and scalable solution for real-time intelligent parking management. This work gives an end-to-end pipeline from annotation conversion to inference and creates a solid basis for further studies in intelligent transport systems.