Dual-branch spatio-temporal graph neural networks for pedestrian trajectory prediction

Xingchen Zhang, Panagiotis Angeloudis, Yiannis Demiris

Abstract: Highlights•A dual-branch Spatio-Temporal GNN for Pedestrian Trajectory Prediction is proposed.•A Finding View Area module is designed to model pedestrian view area automatically.•A Finding Group module is designed to model group behaviors automatically.•Complex social interactions are encoded in adjacency matrices of GNNs.•The proposed method achieves very competitive performance on standard datasets.