VectorNet based prediction of visitor trajectories in heritage sites using VR environmental perception

Taoping Village, a historic Qiang fortress in Sichuan, China, features complex spatial layouts and perception-driven visitor movement, posing challenges for trajectory prediction. This study proposes a VectorNet-based framework integrating VR-based eye-tracking data with vectorized spatial representations. Using data from 45 participants navigating a VR scene of Taoping Village, the model demonstrated improved predictive accuracy (minADE 4.33 m, minFDE 7.19 m) compared to baseline approaches. These results support the feasibility of the approach and its potential integration into behavior-aware Digital Twin systems for heritage site management.