A Unified Object Detection Framework for Front Vehicle Behavior Prediction using YOLOv5s and Ontology Reasoning

In autonomous driving and advanced driver-assistance systems (ADAS), safety and efficiency rely heavily on accurately forecasting front vehicle behavior. This study presents a unified object detection framework that combines ontology-based reasoning with YOLOv5s, a compact and high-performance object recognition model, to improve the prediction of front vehicle behavior. The YOLOv5s model ensures real-time detection of vehicles and traffic objects while providing crucial spatial and contextual information. Complementing this, ontological reasoning leverages domain knowledge—such as traffic laws, vehicle interactions, and environmental factors—to infer high-level behavioral patterns, including lane changes, braking, and turning. By integrating ontological reasoning with the speed and precision of YOLOv5s, the proposed system effectively analyzes vehicle behaviors and predicts potential hazards. Evaluations on real-world driving datasets demonstrate that this unified approach significantly enhances behavior prediction accuracy compared to standalone detection methods. This fusion of machine learning and symbolic reasoning enables autonomous driving algorithms to make more reliable and anticipatory decisions, contributing to safer and more efficient driving systems.