Reliable trajectory prediction in scene fusion based on spatio-temporal Structure Causal Model

Jianmin Liu, Hui Lin, Xiaoding Wang, Lizhao Wu, Sahil Garg, Mohammad Mehedi Hassan

Abstract: Highlights•We reexamine and reinterpret the trajectory prediction task from the perspective of causal reasoning.•We apply the front-door criterion based on causal interventions to ensure that the model can fuse features more fairly.•We construct a dynamic spatio-temporal causal graph to explore dependencies within trajectory.•We utilize counterfactual reasoning representation to enhance the ability to infer future contextual information.•Experiments on two real-world datasets demonstrate that the method exhibits excellent performance and robustness.