YOLO5PKLot: A Parking Lot Detection Network Based on Improved YOLOv5 for Smart Parking Management System

Duy-Linh Nguyen, Xuan-Thuy Vo, Adri Priadana, Kang-Hyun Jo

Abstract: In recent years, the YOLOv5 network architecture has demonstrated excellence in real-time object detection. For the purpose of applying in the smart parking management system, this paper proposes a network based on the improved YOLOv5, named YOLO5PKLot. This network focus on redesigning the backbone network with a combination of the lightweight Ghost Bottleneck and Spatial Pyramid Pooling architectures. In addition, this work also resizes the anchors and adds a detection head to optimize parking detection. The proposed network is trained and evaluated on the Parking Lot dataset. As a result, YOLO5PKLot achieved 99.6% mAP on the valuation set with only fewer network parameters and computational complexity than others.