Interaction-enhanced lane-change early warning in congested traffic: a spatial risk propagation-aware approach

This paper presents an interaction-enhanced Lane-Changing Early Warning (LCEW) system designed to issue reliable early warning signals through a spatial risk-propagation-aware approach. The system is built upon interpretable multi-vehicle trajectory predictions and collision risk identification within an extended LC zone, including multiple downstream and upstream vehicles. We first investigate the stochasticity and hazardousness of LCs, characterized by (i) variable-size multi-vehicle interactions within the LC zone and (ii) the resulting risk propagation from downstream to upstream traffic. To model these stochastic interactions, a Social Spatio-Temporal Graph Convolutional Neural Network framework informed by mutual information (social STGCNN-MI) is introduced to predict multi-vehicle trajectories. By leveraging a mutual information-based kernel function, the framework enhances trajectory prediction accuracy while providing interpretable representations of vehicle interactions. Then, oriented bounding box detection is employed on the predicted trajectories to identify direct collision risks between the LC vehicle and adjacent vehicles and indirect risks among non-adjacent vehicles. These components jointly quantify and characterize how risks originating from downstream interactions influence LC behavior and subsequently propagate to upstream traffic. Finally, awareness of this spatial risk propagation enables the generation of an early warning signal that informs the LC decision-making with potential collision locations within the predicted time window. Traffic simulation experiments conducted in SUMO demonstrate that the proposed LCEW effectively improves vehicle-level safety and comfort, while maintaining system-level traffic safety and efficiency.