The invention provides a deep neural network for vehicle motion trail simulation, the vehicle motion trail is represented by positions of a vehicle at different moments, the deep neural network comprises an attention network, a recurrent neural network and a full-connection neural network, the attention network is used for extracting spatial features from traffic elements, and the recurrent neural network is used for extracting the spatial features from the traffic elements; wherein the traffic elements comprise lane lines, vehicles and pedestrians, and the lane lines, the vehicles and the pedestrians are represented by vectors; the recurrent neural network is used for extracting the state of the current moment according to the spatial features and the state of the previous moment extracted by the recurrent neural network; and the full-connection neural network is used for calculating the position and speed of the vehicle at the next moment according to the state at the current moment. Compared with an existing vehicle motion trail simulation deep neural network, the cross-scene capability and the simulation accuracy of the deep neural network are obviously improved. And the method is also applicable to traffic scenes which are relatively simple than urban intersections.
本发明提供一种用于车辆运动轨迹仿真的深度神经网络,其中车辆运动轨迹以不同时刻车辆的位置表示,深度神经网络包括注意力网络,循环神经网络和全连接神经网络,并且:注意力网络,用于从交通要素中提取空间特征,其中所述交通要素包括车道线,车辆和行人,所述车道线,车辆和行人以向量表示;循环神经网络,用于根据所述空间特征和循环神经网络所提取的前一时刻的状态提取当前时刻的状态;全连接神经网络,用于根据当前时刻的状态计算下一时刻车辆的位置与速度。本发明的深度神经网络相比于现有的车辆运动轨迹仿真的深度神经网络跨场景能力和仿真准确率都有着明显的提升。并且对于比城市路口相对简单的交通场景同样适用。
Urban traffic intersection vehicle motion trail simulation method
一种用于城市交通路口车辆运动轨迹仿真方法
2024-05-31
Patent
Electronic Resource
Chinese
Micro-Simulation Evaluation of Traffic Signal Control in Urban Intersection
British Library Conference Proceedings | 2009
|