Digital Twin-assisted Context-aware Object Detection for Connected and Automated Vehicles

Connected and Automated Vehicles (CAVs) rely heavily on on-board sensing technologies and machine learning (ML) algorithms to perceive the surrounding environment and behave accordingly. Hence, the speed and accuracy of ML models that detect surrounding objects is crucial to ensure road safety. In this work, we propose to ensure detection accuracy while maintaining efficiency by adapting the object detection model to the specific driving context where the CAV is traveling. Fine-tuning has been applied to achieve this objective. With the support of the Vehicular Digital Twin, paired to the CAV, the model that best suits the driving context is identified and loaded into local vehicular computing facilities. Preliminary experimental results highlight the superiority of our proposal compared to the most recent legacy YOLO implementations.