DroneVision: A Low-Code Solution for Urban Parking Occupancy Detection Using Vision-Language Models

Abstract. Accurate detection of on-street parking occupancy is crucial for urban traffic management. We present DroneVision, a low code solution that enables municipalities to analyze parking occupancy using drone-acquired aerial imagery and vision language models (VLM). This approach allows municipal authorities to upload commercial drone imagery and parking space information, automatically processes these inputs, and generates comprehensive analysis reports. We evaluated two state-of-the-art open-source VLMs against traditional pretrained convolutional neural networks (CNNs) for drone imagery analysis. Results demonstrate the effectiveness of our VLM-based approach, offering a scalable solution that can be integrated into smart city infrastructures while making advanced AI technology accessible to municipal authorities through a low-code interface.