<![CDATA[The growing number of vehicles in urban areas is driving the need for smart parking systems capable of automatically detecting parking space availability in real time. This study developed a YOLOv8s-based detection model trained using the PKLot v2 dataset (Roboflow), which consists of a total of 12,416 images from various locations, camera angles, and weather conditions. Preprocessing included converting annotations from the COCO JSON format to YOLO TXT and removing duplicate bounding boxes using an IoU filter of >0.85. The model was trained for 50 epochs using the AdamW optimizer, with mosaic and copy-paste data augmentation, and an image size of 640×640 pixels. Evaluation on the validation subset yielded a mAP@0.5 of 0.9512, precision of 0.9241, recall of 0.9073, and an F1-Score of 0.9156, outperforming similar studies that used YOLOv5 or single-location datasets. A polygon-based Region of Interest (ROI) mechanism was applied to filter out detections outside the parking area, effectively reducing false positives. This study demonstrates that YOLOv8s, optimized on a multi-location dataset, can accurately recognize parking slot availability patterns as the foundation for a smart parking system that can be widely implemented.]]>
Copyrights © 2026
