Urban parking management has become a critical challenge as cities grow more congested. This paper presents a practical drone-based solution using optimized YOLO models to detect parking occupancy in real-time. Unlike traditional fixed cameras, our aerial approach minimizes blind spots while keeping costs low. We evaluate six YOLO variants to find the best speed-accuracy trade-off for parking scenarios, with YOLOv8n emerging as the most efficient for drone deployment. The system goes beyond simple detection—it calculates precise vehicle positions using bounding box centroids and assigns available spaces through Manhattan distance, mimicking actual driving paths. This work bridges computer vision research with practical urban planning needs, offering a scalable path to smarter parking without massive infrastructure investment.