基于图神经网络的交互车辆驾驶意图识别及轨迹预测  

Interactive Vehicle Driving Intention Recognition and Trajectory Prediction Based on Graph Neural Network

作　　者：赵树恩[1] 苏天彬 赵东宇 Zhao Shuen;Su Tianbin;Zhao Dongyu(Chongqing Jiaotong University,Chongqing 400074)

机构地区：[1]重庆交通大学,重庆400074

出　　处：《汽车技术》2023年第7期24-30,共7页Automobile Technology

基　　金：国家自然科学基金项目(52072054);重庆市川渝联合实施重点研发项目(cstc2021jscx-cylh0026);汽车主动安全测试技术重庆市工业和信息化重点实验室开放课题(2021KFKT01)。

摘　　要：为实现周围车辆行驶轨迹的准确预测,运用深度学习方法,设计了一种基于图神经网络与门控循环单元(GRU)的驾驶意图识别及车辆轨迹预测模型。驾驶意图识别模型将车-车间的交互关系构造成时空图,运用图神经网络学习其交互规律,并利用Softmax函数计算出不同驾驶意图的概率;轨迹预测模型采用编码-解码的GRU网络,编码器将车辆历史轨迹信息进行编码并融合识别的驾驶意图信息,再通过解码器实现轨迹预测。最后采用NGSIM数据集对模型进行训练和验证,结果表明:所提出的模型能够更好地识别车辆的驾驶意图,且考虑驾驶意图的车辆轨迹预测模型能够有效提高预测精度。In order to realize the accurate prediction of the trajectory of surrounding vehicles,a driving intention recognition and trajectory prediction model based on graph neural network and Gated Recurrent Unit(GRU)was designed by using deep learning method.The driving intention recognition model constructed the interaction relationship between vehicles into a space-time graph,used the graph neural network to learn its interaction rules and used Softmax function to calculate the probability of different driving intentions.The trajectory prediction model adopted an encoded-decoded GRU network and the encoder encoded the vehicle history trajectory information and fused the recognized driving intention information and then realized trajectory prediction through the decoder.Finally,the Next Generation Simulation(NGSIM)dataset was used to train and verify the model and the results show that the proposed model can better identify the driving intention of the vehicle,and the vehicle trajectory prediction model considering the driving intention can effectively improve the prediction accuracy.

关 键 词：自动驾驶 驾驶意图识别 轨迹预测 图神经网络 门控循环单元 

分 类 号：U461.91[机械工程—车辆工程]