The present invention relates to a method and system for modeling a spatio-temporal diagram of road users in an observation frame of an environment in which an autonomous vehicle is operating (i.e., a traffic scene), clustering the road users into several categories, and providing the spatio-temporal diagram to a trained graphical convolutional neural network, gNN) to predict future pedestrian actions. The future pedestrian action may be one of that the pedestrian will pass through the roadway and that the pedestrian will not pass through the roadway. The spatiotemporal diagram includes a better understanding of the observation frame (i.e., a traffic scene).
本发明涉及用于以下操作的方法和系统:在自动驾驶车辆运行的环境的观察帧(即交通场景)中对道路使用者进行时空图建模,将所述道路使用者聚类为几种类别,并将所述时空图提供给经过训练的图形卷积神经网络(graphical neural network,GNN)以预测未来的行人动作。所述未来的行人动作可以是:行人将穿过道路和行人将不穿过道路中的一个。所述时空图包括对所述观察帧(即交通场景)的更好理解。
Method and system for pedestrian motion prediction based on graph neural network in automatic driving system
用于自动驾驶系统中基于图神经网络的行人动作预测的方法和系统
2024-01-23
Patent
Elektronische Ressource
Chinesisch
Europäisches Patentamt | 2023
|Europäisches Patentamt | 2022
|Pedestrian trajectory prediction method and automatic driving method of vehicle
Europäisches Patentamt | 2023
|