PTP-STGCN: Pedestrian Trajectory Prediction Based on a Spatio-temporal Graph Convolutional Neural Network

It is the prerequisite to ensure the safety of road users in traffic scenes for the application of autonomous vehicles. Pedestrians are the main participants in traffic scenes, and reasonable inference and prediction of their future trajectories are crucial for autonomous driving technology and road safety. Pedestrian trajectories are highly dynamic, apparently random, and complex to interact with traffic environment agents; therefore, selective modeling of spatial interaction and temporal dependence of pedestrians is necessary. To address this challenge, this paper proposes a novel pedestrian trajectory prediction model based on a spatio-temporal graph convolutional network (PTP-STGCN). Specifically, a new crowd interaction attention (CIA) function is defined to quantify the interaction information between adjacent pedestrians better. This function captures the spatial interaction features of pedestrians at each time step by a spatial graph convolution network (S-GCN). Meanwhile, the time-series motion features of each pedestrian are extracted by a temporal transformer network (T-transformer), and a spatio-temporal interaction graph of pedestrian features is constructed by the STGCN composed of the S-GCN and T-transformer. Finally, a time-extrapolator convolutional neural network (TXP-CNN) is used to predict pedestrian trajectories in the time dimension of the STGCN. The experimental results on the ETH and UCY datasets show that the proposed model achieves better performance than the state-of-the-art baselines regarding the average displacement error (ADE) and final displacement error (FDE). Due to the excellent performance in pedestrian trajectory prediction, the proposed network achieves more predictive final planned trajectory of an autonomous vehicle, while avoiding unnecessary trajectory changes and collision risk, thus providing a promising solution for practical pedestrian trajectory prediction applications.