Social Spatio-Temporal Graph Convolutional Neural Network for Pedestrian Trajectory Prediction

Pedestrian trajectory prediction serves as a critical component in autonomous vehicle perception technology, playing a vital role in safe navigation path planning. The complex interaction behaviors among surrounding traffic participants significantly influence pedestrian movement patterns, making accurate trajectory prediction particularly challenging. We propose a Social Spatial-Temporal Graph Convolutional Neural Network (Social-STGCNN) that models crowd interactions through graph structural representations rather than conventional aggregation approaches. Experimental evaluation on an urban intersection dataset from Nanjing demonstrates our model achieves a final displacement error (FDE) of 0.52 and an average displacement error (ADE) of 0.33.Quantitative analysis shows that the accuracy of the model has been improved. In addition, this model has data efficiency and surpasses the advanced level of previous ADE metrics with only 20% of training data. The model also includes a kernel function that embeds social interactions between pedestrians into the adjacency matrix.