This paper presents a domain-specific object detection pipeline for a virtualized campus environment, fine-tuning a YOLOv8 model to recognize chairs, benches, cars, and bicycles. By combining targeted data collection, manual annotation, and iterative training on GPU hardware, we obtain an accurate and competitive detector using a moderate dataset. Evaluation on image sequences derived from the digital twin indicates advantages over a general-purpose baseline for the intended classes. This work highlights the effectiveness of specialization in object detection tasks and underscores the practical advantages of integrating computer vision with spatial digital twin frameworks.