智能驾驶系统对周边交通车辆预期轨迹的单模态网络预测方法  

A Uni-Modal Network Prediction Method for Surrounding Vehicle Expected Trajectory in Intelligent Driving System

作　　者：高振海[1] 鲍明喜 高菲[1] 唐明弘 吕颖 Gao Zhenhai;Bao Mingxi;Gao Fei;Tang Minghong;LüYing(State Key Laboratory of Automotive Simulation and Control,Jilin University,Changchun,130000;General Research and Development Institute,China FAW Corporation Limited,Changchun 130013;State Key Laboratory of Comprehensive Technology on Automobile Vibration and Noise&Safety Control,Changchun 130013)

机构地区：[1]吉林大学,汽车仿真与控制国家重点实验室,长春130000 [2]中国第一汽车股份有限公司研发总院,长春130013 [3]汽车振动噪声与安全控制综合技术国家重点实验室,长春130013

出　　处：《汽车技术》2022年第11期1-9,共9页Automobile Technology

基　　金：国家自然科学基金项目(51775236,51675224,U1564214)。

摘　　要：为提升自动驾驶汽车准确预测周围车辆驾驶行为及轨迹的能力,提出一种基于单双向长短时记忆(MB-LSTM)的行为意图识别及交通车辆轨迹预测模型。该模型中行为意图识别模块输出被预测车辆车道保持、左换道、右换道、加速左换道和加速右换道的概率;交通车辆轨迹预测模块结合上下文向量和行为意图信息预测未来坐标和速度信息。通过HighD数据集对模型进行训练、验证与测试。验证结果表明:基于环境交互信息构建的车辆预期轨迹预测模型在预测长时域轨迹时具有较高的精度。In order to improve the ability of autonomous vehicles to accurately predict the driving behavior and trajectories of surrounding vehicles, the paper proposes a behavioral intent recognition and traffic vehicle trajectory prediction model based on Monodirectional and Bidirectional LSTM(MB-LSTM), in which the behavioral intention recognition module outputs the probability of the predicted vehicles in lane keeping, left lane changing, right lane changing,accelerating left lane changing and accelerating right lane changing;the traffic vehicle trajectory prediction module combines context vectors and behavioral intention to predict future information of coordinates and speed. The model is trained, validated and tested by the HighD dataset. The experimental results show that the expected trajectory prediction model constructed based on environmental interaction information has high accuracy in predicting long-term domain trajectories.

关 键 词：轨迹预测 行为意图识别 交互式行为 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] U463.6[自动化与计算机技术—计算机科学与技术]