Towards V2X-Enabled Vulnerable Road Users Protection: Baseline Performance Evaluation of a Fundamental Kinematic Model for CAM-Based Trajectory Prediction at Road Intersections

Enhancing the safety of Vulnerable Road Users (VRUs) at urban intersections is a critical challenge for Intelligent Transportation Systems (ITS). This paper presents a baseline performance evaluation of a fundamental kinematic model for Vehicle-to-Everything (V2X) trajectory prediction. Using a co-simulation framework of SUMO, OMNeT++, and Artery, we evaluate a linear regression model's performance across two Vehicle-to-Pedestrian (V2P) scenarios with varying pedestrian dynamics (walking vs. running) and two prediction horizons (1s and 2 s). Our results reveal a significant “predictability gap” whose nature depends on the prediction horizon. For short-term (1s) predictions, the model is surprisingly more accurate for walking pedestrians than for vehicles. However, the model's accuracy for vehicles degrades exponentially at the longer (2s) prediction horizon, in contrast to the pedestrian prediction error, which degrades more gradually. We also find that a higher data frequency from a more dynamic road user is insufficient to improve prediction accuracy. This study validates the distinct failure modes of simple kinematic models for different road users and horizons, establishing a solid performance baseline that motivates the development of more sophisticated, context-aware algorithms.