汽车工程 ›› 2022, Vol. 44 ›› Issue (11): 1647-1655.doi: 10.19562/j.chinasae.qcgc.2022.11.003

所属专题: 智能网联汽车技术专题-感知&HMI&测评2022年

• • 上一篇    下一篇

自动驾驶汽车测试场景基元自动提取方法

朱冰,孙宇航,赵健(),张培兴,范天昕,宋东鉴   

  1. 吉林大学,汽车仿真与控制国家重点实验室,长春  130022
  • 收稿日期:2022-04-28 修回日期:2022-05-31 出版日期:2022-11-25 发布日期:2022-11-19
  • 通讯作者: 赵健 E-mail:zhaojian@jlu.edu.cn
  • 基金资助:
    国家自然科学基金(52172386);工信部产业技术基础公共服务平台项目(2020-0100-4-1);长沙市“揭榜挂帅”重大科技项目(kq2102008)

Automatic Extraction Method for Autonomous Vehicle Test Scene Primitives

Bing Zhu,Yuhang Sun,Jian Zhao(),Peixing Zhang,Tianxin Fan,Dongjian Song   

  1. Jilin University,State Key Laboratory of Automotive Simulation and Control,Changchun  130022
  • Received:2022-04-28 Revised:2022-05-31 Online:2022-11-25 Published:2022-11-19
  • Contact: Jian Zhao E-mail:zhaojian@jlu.edu.cn

摘要:

本文中提出了一种基于自然驾驶数据库的自动驾驶汽车测试场景基元自动提取方法。以隐马尔科夫模型为框架,采用向量自回归模型为观测概率分布函数,采用层次狄利克雷过程为模型进行先验分布和后验更新,通过解耦过程和黏性过程抑制模型隐状态的快速切换,实现给定观测数据下隐状态的求解,并根据隐状态对场景基元进行自动划分。最后,以一段随机自然驾驶序列为例进行了算法测试验证,结果表明,所提出的方法可从自然驾驶数据库中非参数地、可解释地、全自动地提取测试场景基元,所提取的场景基元具有明确的物理意义,可为基于场景的自动驾驶汽车测试奠定良好基础。

关键词: 自动驾驶汽车测试, 场景基元提取, 自然驾驶数据库, 隐马尔可夫模型

Abstract:

In this paper, an automatic test scene primitive extraction method from the naturalistic driving dataset for autonomous vehicles is proposed. With the hidden Markov model (HMM) as framework, the extraction method uses the vector autoregressive model as the observation probability distribution function, and adopt the Hierarchical Dirichlet Process to fulfill prior distribution and posterior updating for HMM. Then the disentangled and sticky process are used to suppress the frequent switching of the HMM hidden states for achieving the solution of hidden states with given observation data, and the scene primitives are automatically divided according to the hidden states. Finally, a piece of naturalistic driving series is randomly selected to verify the method. The results show that the method proposed can automatically extract explicable test scene primitives with clear physical meanings from the naturalistic driving dataset without setting parameters, laying a good foundation for scene-based autonomous vehicle testing.

Key words: autonomous vehicle testing, scene primitives extraction, naturalistic driving dataset, hidden Markov model