A Set of Control Points Conditioned Pedestrian Trajectory Prediction

Predicting the trajectories of pedestrians in crowded conditions is an important task for applications like autonomous navigation systems. Previous studies have tackled this problem using two strategies. They (1) infer all future steps recursively, or (2) predict potential destinations at once and interpolate intermediate to arrive there. However, these strategies often suffer from accumulated errors recursive inference, restrictive assumptions about social relations path. In paper, we present a graph convolutional network-based trajectory prediction. Firstly, propose control point prediction that divides path into three sections infers reduce error. To do this, construct multi-relational weighted graphs account their physical complex relations. We then introduce refinement step based on spatio-temporal graph. By considering interactions between neighbors, better results are achievable. experiments, proposed network achieves state-of-the-art performance various real-world benchmarks.