IST-PTEPN: an improved pedestrian trajectory and endpoint prediction network based on spatio-temporal information

The prediction of pedestrian trajectories in complicated dynamic situations has garnered a great deal of interest among researchers and academics, and it plays a crucial role in numerous domains, including autonomous vehicles, intelligent robotics, and video surveillance. In this study, we offer the IST-PTEPN, a trainable and interpretable end-to-end model for predicting pedestrian trajectory. IST-PTEPN encodes the spatial and temporal characteristics of pedestrian trajectories and surrounding scenes with CNN and Transformer, then feeds the encoded vectors into Endpoint Classify CNN to generate predicted endpoints of the trajectories, and finally combines TCN and GAN to generate high-quality pedestrian trajectories. Experiments on two public datasets, ETH and UCY, demonstrate that our IST-PTEPN pedestrian trajectory prediction and endpoint prediction method outperforms the mainstream state-of-the-art methods.