EfficientDet-Based Event Detection for Smart Parking System using Ontology and SPARQL Query

Currently, the dynamic nature of traffic management demands mobile surveillance capabilities, which are practical for the integration of 5G networks and edge computing advancements. The advancement of Multi-access Edge Computing (MEC) has enabled data processing closer to the source, significantly reducing latency and enhancing responsiveness. Cities require dynamic and difficult monitoring strategies to identify illegally parked vehicles effectively. Mobile surveillance is effectively carried out using platforms such as Unmanned Aerial Vehicles (UAVs) or ground vehicles, which gather multimodal data and transmit it to the closest edge server through a 5G connection for rapid processing. The edge server transmits actionable insights to the control unit with minimal delay, allowing an effective arrangement of real-time reduction strategies. Vehicles equipped with cameras and sensors traverse urban areas, capture roadside data, and transmit it to nearby edge servers through a 5G network for real-time analysis. EfficientDet is employed as the main object detection model to accurately localize vehicles within the video frames. Ontology-based reasoning permits surveillance data and facilitates the construction of knowledge graphs. Semantic queries are stated using SPARQL, and Description Logics (DL) allows recapture information without the need to store full video sequences. The proposed EfficientDet-Based Smart Parking System provides a better accuracy of 98.4% than the existing Onto-ViBeNet (Ontology-enhanced Vehicle Behavior Prediction with YOLOv5s).