A Vehicle Trajectory Prediction Model for Highways Based on Multi-modal Feature Fusion

For complex highway traffic scenarios, this paper proposes a vehicle trajectory prediction model that integrates multi-modal features with physical constraints. First, the model employs a highway map generator to extract structured road features such as lane markings and road boundaries. Subsequently, a multi-modal feature fusion encoding module is constructed, integrating a map encoder, a spatial interaction encoder, a temporal dynamics encoder, a driving intention recognition module, and a feature fusion module to deeply fuse road prior information with dynamic interaction features among vehicles. Finally, a trajectory decoding module based on multi-head attention mechanism (MHAM) and a gated recurrent unit (GRU) is designed to decode the future trajectory sequence. A physical constraint layer is embedded within this module to enforce realistic bounds on vehicle motion, guaranteeing physically plausible trajectories. Experimental results on the HighD dataset demonstrate that our approach significantly outperforms mainstream baseline algorithms, achieving considerable improvements in both short-Term and long-Term prediction accuracy. The proposed model shows strong capability in handling complex highway driving scenarios.