无人驾驶车辆城市交叉口周边车辆轨迹预测  

Trajectory Prediction of Surrounding Vehicles for Unmanned Vehicle at Urban Intersections

作　　者：陈雪梅[1] 李梦溪 王子嘉 欧洋佳欣 CHEN Xuemei;LI Mengxi;WANG Zijia;OUYANG Jiaxin(School of Mechanical Engineering,Beijing Institute of Technology,Beijing 100081,China)

机构地区：[1]北京理工大学机械与车辆学院,北京100081

出　　处：《汽车工程学报》2021年第4期235-242,共8页Chinese Journal of Automotive Engineering

基　　金：国家自然科学基金青年科学基金(51705021);国家自然科学基金汽车产业联合基金(U1764261);国家自然科学基金重点项目(91420203)。

摘　　要：针对城市交叉口周边车辆长时轨迹预测问题,搭建路基和实车采集平台采集大量轨迹数据,采用高斯混合模型(Gaussian Mixture Model,GMM)识别目标运动模式,采用高斯过程回归(Gaussian Processes Regression,GPR)模型进行城市交叉口周边车辆轨迹长时预测,采用路基数据集对预测模型进行交叉验证。针对实车场景,提出结合无迹卡尔曼滤波的高斯过程算法(GP-UKF),并采用实车数据对该算法进行离线测试。结果表明,GMM可以有效提取车辆运动模式,GPR模型在长时轨迹预测问题上的表现优于基于物理模型的预测算法,并且GP-UKF模型对目标的长时轨迹预测具有更高的精度。In order to solve the problem of long-term vehicle trajectory prediction at urban intersections,the paper built a platform that collects a large amount of trajectory data from the real vehicle and subgrades.The Gaussian mixture model was adopted to train the model of motion pattern recognition.And the Gauss process regression model was used to predict the long-term trajectory of surrounding vehicles.The roadbed datasets were used for cross-validation of the prediction model.The GP algorithm combined with unscented Kalman algorithm was proposed and tested offline with real vehicle data.The results prove the effectiveness of the motion mode recognition model.The GPR algorithm outperforms other physical model-based algorithm in long-term trajectory prediction.And the GP-UKF model combining with the filtering algorithm has better predictive performance than the GPR algorithm.

关 键 词：无人驾驶 轨迹预测 高斯混合模型 高斯过程回归 城市交叉口 

分 类 号：U461.99[机械工程—车辆工程]