Abstract:

In order to predict the trajectory of pedestrians in complex scenes, an interpretable model based on generative adversarial networks is proposed. The model takes the historical trajectory information of the pedestrians in the scene and the environment information of the scene as the input of the model, and introduces physical attention mechanism and social attention mechanism into the generative adversarial network to predict the pedestrian trajectory. Among them, the physical attention mechanism helps to model the overall layout of complex scenes and extract significant features related to paths in the image. The social attention mechanism can model the impact of different pedestrian interactions on future trajectories. Under the overall framework of generative adversarial networks, the combination of physical and social attention mechanisms enables the model to predict multiple acceptable future paths that meet physical constraints and social behavior norms. Experiments on simulation data and real standard datasets prove the model can effectively predict the future trajectory of pedestrians.

Key words: trajectory prediction, generative adversarial network, attention mechanism

摘要：

为了预测行人在复杂场景中的行走轨迹，提出了一种基于生成对抗网络的可解释模型。该模型以场景中行人的历史轨迹信息和场景环境信息作为模型的输入，并在生成对抗网络中引入了物理注意力机制和社会注意力机制对行人轨迹进行预测。其中，物理注意力机制有助于建模复杂场景的整体布局并提取图像中与路径相关的显著特征，社会注意力机制能够建模不同行人之间的交互对未来轨迹的影响。在生成对抗网络的整体框架下，物理和社会注意力机制的结合使得该模型能够预测出符合物理限制和社会行为规范的多条可接受的未来路径。通过在仿真数据和真实的标准数据集上的实验，可以证明该模型能够实现对行人未来轨迹的有效预测。

关键词: 轨迹预测, 生成对抗网络, 注意力机制