Cognitive Internet of Vulnerable Road Users in Traffic: Predictive Neural Modulations of Road Crossing Intention

Vulnerable Road Users (VRUs) present a significant challenge for road safety due to the frequent unpredictability of their behaviors. In typical Intelligent Transportation Systems, vision-based approaches supported by networked cameras are often used to anticipate VRUs motion intentions and trajectories. However, several limitations posed by occlusions and distractions set a boundary for the efficacy of such methods. To address these challenges, this study introduces a framework that leverages data collected using wearable neurophysiological sensors on VRUs to integrate them seamlessly into the Vehicle-to-Everything communication framework. This integration empowers VRUs to autonomously broadcast their intended movements to other road agents, especially autonomous vehicles, thereby bridging a critical gap in current vehicular communication systems. To validate this concept, we conducted an experiment involving 12 participants, from whom EEG signals were collected as they engaged in road-crossing decisions within simulated environments. Employing Hidden Markov Models, we identified four cognitive stages intrinsic to a pedestrian's decision-making process. Our statistical analysis further revealed significant variations in EEG activities across these stages, shedding light on the neural correlates and cognitive dynamics underpinning pedestrian road-crossing behavior. We then developed a predictive cognitive model using dynamic time warping and K-nearest neighbors algorithms, optimized through a data-driven sliding window approach. This model demonstrated high predictive accuracy, evidenced by an Area Under the Curve of 0.91, indicating its capability to anticipate pedestrian road-crossing actions approximately 1 second in advance of any pedestrian movement. This research paves the way for a novel VRU-Vehicle interaction paradigm and signifies a shift towards a forward-thinking ecosystem.