基于Bi-GLSTM网络的车辆驾驶意图分析与识别

Driving Intention Recognition Model Based on Bi-GLSTM Network

驾驶意图识别能有效提高自车对其他交通参与者的轨迹预测能力,是实现智能车自主决策和规划的基础。然而动态复杂交通环境下周围车辆的交互是实现准确可靠驾驶意图识别亟待解决的挑战之一。为提高在动态复杂交通场景下驾驶意图识别的准确率,提出基于双向图长短时记忆网络(Bidirectional graph long short term memory,Bi-GLSTM)网络的驾驶意图时序识别模型。首先基于局部加权回归散点平滑法对原始数据集中的位置、速度和加速度进行平滑处理,并联合纵横向运动参数为数据标注驾驶意图;然后建立图注意力神经网络,分析和提取周围车辆与目标车辆之间的交互特征,嵌入注意力机制,分析周围车辆对目标驾驶意图的重要性,增强模型对相关性较大的车辆运动状态关注程度;融合周围车辆交互特征和目标车辆历史运动特征,为提高模型在动态复杂交通环境下的鲁棒性和可靠性,基于双向长短时记忆网络提取特征之间的时序特征;最后在公开数据集HighD上训练并验证模型的有效性,结果表明相比于图神经网络、循环神经网络等模型,识别准确率分别提高了11.33%、55.31%;通过可视化注意力权重,说明所提出的模型也一定程度上解决了可解释性问题。

Driving intention recognition is promised to effectively improve the vehicle’s ability to predict the trajectory of other traffic participants.However,the interaction of surrounding vehicles in a dynamic and complex traffic environment is one of the most challenges to be solved.In order to improve the accuracy of driving intention recognition in dynamic and complex traffics,a time-series recognition model of driving intention based on Bi-GLSTM network is proposed.First,the position,velocity and acceleration in the original dataset are smoothed based on the Local weighted regression scatter smoothing method.Also,the driving datasets are labeled with driving intention based on the longitudinal and lateral motion parameters.Subsequently,a graph attention neural network is established to extract interaction features among surrounding vehicles,where attention mechanism is embedded,to enhance highly related vehicle motion states.To further improve the robustness of the model in dynamic and complex traffics,Bidirectional long short-term memory network is used to extract deep temporal features of interaction and historical motion.Moreover,our model is trained and verified on the public datasets HighD.Compared with GNN and RNN model,the recognition accuracy increased by 11.33%,55.31%.By visualizing the attention weight,it shows that the proposed model also solves the problem of explainability to a certain extent.