The invention discloses a trajectory prediction method for vehicles around an automatic driving vehicle in a hybrid driving scene, which comprises the following steps: firstly, aggregating vehicle trajectory data sensed by a single automatic driving vehicle by using GNN, constructing a dual graph to enhance the representation of interaction between vehicles, and then providing an attention intermediate feature fusion network, the method globally covers the track features obtained by a plurality of autonomous vehicles, and improves the accuracy of a track prediction result. The method can be more suitable for a real traffic scene supporting a multi-vehicle cooperation technology, can better fuse the intermediate features of a plurality of autonomous vehicles, and can more accurately predict the future trajectory of the vehicle.
本发明公开了一种混合驾驶场景中自动驾驶车辆周围车辆的轨迹预测方法,首先使用GNN聚合单个自动驾驶车辆感知的车辆轨迹数据,同时构建了对偶图用来增强车辆间交互作用的表示,然后提出一种注意力中间特征融合网络,全局覆盖多个自动驾驶车辆获取的轨迹特征,提升轨迹预测结果的精准性。本发明能够更加适用于支持多车协同技术的真实交通场景,能更好地融合多个自动驾驶车辆间的中间特征,能够更加准确地预测车辆未来轨迹。
Trajectory prediction method for vehicles around autonomous vehicle in hybrid driving scene
混合驾驶场景中自动驾驶车辆周围车辆的轨迹预测方法
2024-02-06
Patent
Electronic Resource
Chinese
Driving scene based path planning for autonomous driving vehicles
European Patent Office | 2021
|