A vehicle motion prediction method includes the following steps. And generating a road graph structure according to the original map data. And according to the track data of the plurality of vehicles in the plurality of continuous frames and the road graph structure, establishing a plurality of placeholder flow graphs having the same data format as the road graph structure. And according to the trajectory data of the plurality of vehicles in the plurality of continuous frames, connecting a plurality of time edges among the plurality of placeholder flow diagrams to establish a time placeholder flow diagram. Feature aggregation is performed on the time placeholder map to generate a plurality of updated node features, and a motion prediction of the self-driving vehicle is generated based on the plurality of updated node features. When feature aggregation is carried out on the time occupancy flow graph, the vehicle-to-vehicle interaction relationship and the vehicle-to-road interaction relationship can be extracted at the same time, so that information deviation caused by extraction of the two interaction relationships at different attention layers is avoided, and the accuracy of vehicle motion prediction of the self-driving vehicle is improved.
一种车辆运动预测方法,包含下列步骤。根据原始地图数据,产生道路图形结构。根据多个车辆在多个连续帧幅中的轨迹数据以及道路图形结构,建立与道路图形结构具有同质的数据格式的多个占位流图。根据所述多个车辆在所述多个连续帧幅中的轨迹数据,将多个时间边缘连接在所述多个占位流图之间,以建立时间占位流图。对时间占位流图进行特征聚合以产生多个更新后节点特征,且根据所述多个更新后节点特征产生自驾车辆的运动预测。在对时间占位流图进行特征聚合时,能同时提取车辆对车辆交互关系以及车辆对道路的交互关系,从而避免前述两种交互关系在不同注意力层提取所造成的信息偏差,以提高自驾车辆的车辆运动预测的准确性。
Vehicle motion prediction method and calculation system
车辆运动预测的方法及计算系统
2024-11-22
Patent
Electronic Resource
Chinese
IPC: | G08G Anlagen zur Steuerung, Regelung oder Überwachung des Verkehrs , TRAFFIC CONTROL SYSTEMS |
OTHER VEHICLE MOTION PREDICTION METHOD AND OTHER VEHICLE MOTION PREDICTION DEVICE
European Patent Office | 2021
|OTHER VEHICLE MOTION PREDICTION METHOD AND OTHER VEHICLE MOTION PREDICTION DEVICE
European Patent Office | 2020
|