Uncertainty and Prioritization: Empirical Evaluation of a VoI-Based CPM Generation Pipeline Using Real-World Data

Modern Intelligent Transportation Systems (ITSs) rely on accurate and timely data exchange over vehicular communication networks to ensure road safety and efficiency. However, current ETSI dynamic-based message generation rules for Collective Perception Messages (CPMs) exhibit limitations under real-world sensor inaccuracies, often resulting in excessive bandwidth usage or transmission of noisy object states. In this paper, we introduce a novel Value-of-Information (VoI)-based CPM generation pipeline that systematically prioritizes the transmission of high-quality object data. Different information theory-based VoI approaches are investigated, namely Kullback-Leibler Divergence (KLD) and Expected Covariance Divergence (ECD). Unlike previous approaches, our pipeline integrates a track association and fusion procedure that accounts for multi-source measurements, a VoI-divergence threshold to filter low-value information, and a VOI-based resource allocation mechanism that ensures scalability under congested conditions. We evaluate the approach on a real-world intersection trajectory dataset with diverse traffic participants, ranging from vehicles to Vulnerable Road Users (VRUs), and demonstrate significant improvements in object-tracking accuracy and reduced propagation of inaccurate information. Despite transmitting at bandwidth levels comparable to the baseline, our VoI-based system delivers more consistent and precise CPM updates. Notably, our findings reveal that baseline ETSI rules generate larger CPMs with increasing sensor inaccuracy, potentially harming scalability in dense networks. By contrast, our VoI-based pipeline suppresses noisy, unreliable data, leading to highly reduced dissemination of inaccurate information.