Social-Interaction GAN: Pedestrian Trajectory Prediction.

With the increasing number of intelligent autonomous systems in human society, the ability of such systems to perceive, understand and anticipate human behaviors becomes increasingly important. However, the pedestrian trajectory prediction is challenging due to the variability of pedestrian movement. In this paper, we tackle the problem with a deep learning framework by applying a generative adversarial network (GAN) and introduce a model called Social-Interaction GAN (SIGAN). Specially, we propose a novel Social Interaction Module (SIM) to dispose the human-human interactions, which combines the location and velocity features of the pedestrians in a local area. Extensive experiments show that our proposed model can obtain state-of-the-art accuracy.