Vision-Based Motorcycle Detection for Parking Density Estimation: A YOLOv8 Approach with System Interface Simulation

Efficient monitoring of parking availability is essential for reducing search time, congestion, and inefficient vehicle mobility within parking facilities. This study proposes a vision-based parking density estimation system for motorcycle parking areas using the YOLOv8 object detection model. Rather than performing exact per-slot occupancy classification, parking conditions are represented through vehicle-count-based density estimation, which is more robust under real parking conditions characterized by occlusions and irregular parking patterns. The model is trained and evaluated on a custom motorcycle parking dataset, with hyperparameter experiments conducted using three optimizers (Adam, SGD, and RMSProp) and three batch sizes (8,16, and 32) to identify the optimal configuration. Experimental results show that the Adam optimizer with a batch size of 16 achieves the best performance, yielding an mAP50 of 0.978 and an mAP50-95 of 0.681. These results indicate strong detection accuracy and reliable counting performance for parking density analysis. A simple interface simulation is further implemented to demonstrate how the estimated parking density can be used to recommend less congested parking areas.