| 研究実績の概要 |
A reliable pedestrian trajectory forecasting model in crowds plays a critical role in safety driving of autonomous vehicles. With the success of deep learning, researches are burgeoned to address the problems, but they ignore the interpretability of social interactions and multi-modalities. Besides, generalization of prediction models to unseen scenarios and objects is important to real-world applications, but rarely researched. This research proposes to address the above problems and estimates reasonable pedestrian future trajectories by interpreting and incorporating heterogeneous interacting motion sequences jointly, revealing explainable multi-modal probabilistic distributions of future trajectories and considering the ability to generalize well to unseen scenarios and objects.
|
