A Pedestrian Trajectory Prediction Model Based on Interactive Attention Mechanism

The interaction between the pedestrian and the other pedestrians in a shorter period of time has a great influence on the trajectory of the pedestrian. The Social Interaction-Generative Adversarial Network (SI-GAN), a new method based on Social-Generative Adversarial Network (S-GAN), is proposed in this research in an effort to address the current scenario where the pedestrian trajectory prediction method does not fully utilize the pedestrian interaction information. Between the Generator’s Encoder and Decoder is put the interactive attention module The 4 types of interaction information are used by the attention mechanism to calculate the influence weights of pedestrians. In order to create a pedestrian trajectory consistent with social norms, the obtained weights are fed into the Decoder together with information about the pedestrian’s history trajectory information. The Discriminator then decides if the trajectory is a real trajectory or a Generator-generated trajectory. The results shows that the SI-GAN model’s Average Displacement Error (ADE) and Final Displacement Error (FDE) are both decreased when compared to the S-GAN model by 11.38% and 17.90%, respectively.