A physics-guided and adaptive fusion framework for vehicle trajectory reconstruction on freeways

In mixed traffic environment composed of regular vehicles (RVs), probe vehicles (PVs), and connected automated vehicles (CAVs), the low penetration rates of PVs and CAVs pose significant challenges in capturing comprehensive traffic flow characteristics. This study proposes a novel framework for vehicle trajectory reconstruction on freeways, which integrates the Intelligent Driver Model (IDM) with a Long Short-Term Memory (LSTM) network through a physics-guided and adaptive fusion module. The fusion module incorporates physical constraints, including overtaking behaviour and speed contour map. The IDM and LSTM models are then employed to reconstruct the missing RV trajectories, and their outputs are integrated through the fusion module. Finally, the fused trajectories are smoothed using a Savitzky – Golay filter. The proposed method is evaluated using the NGSIM dataset under various reconstruction scenarios. Experimental results demonstrate that the proposed framework significantly outperforms baseline models in terms of trajectory reconstruction accuracy.