Pedestrian trajectory prediction based on pedestrian velocity threshold constraints

How to accurately predict pedestrian trajectories in multiple real-world scenarios is crucial for autonomous driving. In this paper, we propose a physical constraint module based on pedestrian velocity thresholds to improve the unreasonable trajectories predicted by the baseline model, while we introduce a new velocity discrepancy loss function to more efficiently inversely update the network weights while establishing the correlation between the pedestrian motion speeds in the historical time and the future time. Our proposed method is a plug-and-play module, and we evaluate the effectiveness of our method on the ETH and UCY datasets using two different baseline models.