Pedestrian Trajectory Prediction Based on Social Pooling and Spatio-Temporal Graph Convolution Network

In recent years, trajectory prediction has become a major focus in the field of smart cities. This paper investigates the use of Socially Acceptable Trajectories with Generative Adversarial Networks (Social GAN) for pedestrian trajectory prediction, specifically addressing the issue of social interactions between pedestrians. We optimized the Socially Acceptable Trajectories with Generative Adversarial Networks for this purpose. In our study, we preprocessed real-world pedestrian data, taking into account factors such as pedestrian interactions, the dimensions of graph expansion, and the distribution of each neural network layer, to enhance accuracy in trajectory prediction. We introduced three social application methods based on the social pooling layer, converting the Social GAN into a Spatio-Temporal Graph Convolution Network (ST-GCN) which led to improved performance in terms of Average Displacement Error (ADE) and Final Displacement Error (FDE). Our experimental results demonstrate that our proposed method shows significant improvement in both ADE and FDE compared to traditional GANs.