Integrasi YOLOv11 dan Intersection-Based Method Untuk Estimasi Karakteristik Parkir Berdasarkan Parking Lot Surveillance Video

The rapid growth of vehicles without a corresponding increase in parking space availability has led to various issues such as traffic congestion, fuel waste, and excessive emissions. This study develops a computer vision-based parking analysis system using the YOLOv11 model to automatically detect vehicles in parking areas. The system integrates an intersection-based method and the BoT-SORT object tracking algorithm to classify parking spot availability. The classification results are then used to extract parking characteristic data. Video data were obtained from a publicly accessible livestream on YouTube in Kusatsu, Japan, and used for training and evaluating the model. The model achieved an mAP@50-95 of 0.926 under bright lighting conditions and 0.859 in low-light conditions. Additionally, estimation accuracy was evaluated using MAE and R² metrics, showing promising results, with MAE of 1.27 and R² of 0.989 during daytime, and MAE of 0.91 and R² of 0.91 at night.