欢迎访问《汽车安全与节能学报》，
汽车安全与节能学报

JASE ›› 2019, Vol. 10 ›› Issue (4): 391-412.DOI: 10.3969/j.issn.1674-8484.2019.04.001

• 综述与展望 •    下一篇

智能汽车决策中的驾驶行为语义解析关键技术

李国法 1，陈耀昱 1，吕 辰 2，陶 达 1，曹东璞 3，成 波4   

  1. （1. 深圳大学 机电与控制工程学院人因工程研究所， 深圳 518060，中国；
    2. 南洋理工大学 机械与航空航天工程学院， 639798， 新加坡； 
    3. 滑铁卢大学 机械与机电工程系， 滑铁卢 N2L3G1，加拿大;
    4. 汽车安全与节能国家重点实验室， 清华大学， 北京 100084，中国)
  • 收稿日期:2019-10-31 出版日期:2019-12-31 发布日期:2020-01-01
  • 作者简介:李国法 (1986 －)，男( 汉)，河南，特聘副研究员。E-mail: guofali@szu.edu.cn。
  • 基金资助:
     国家自然科学基金青年项目(51805332) ；第 4 届中国汽车工程学会青年人才托举工程培养计划；广东省自然 科学基金 (2018A030310532)；汽车安全与节能国家重点实验室开放基金 (KF1801)。

Key techniques of semantic analysis of driving behavior in decision making of autonomous vehicles

LI Guofa 1，CHEN Yaoyu1，LÜ Chen2，TAO Da1，CAO Dongpu3，CHENG Bo4   

  1. (1. Institute of Human Factors and Ergonomics, College of Mechatronics and Control Engineering, Shenzhen University, Shenzhen 518060, China; 
    2. School of Mechanical and Aerospace Engineering, Nanyang Technology University, 639798, Singapore；
    3. Department of Mechanical and Mechatronics Engineering, University of Waterloo, N2L3G1, Canada；
    4. State Key Laboratory of Automotive Safety and Energy, Tsinghua University, Beijing 100084, China）
  • Received:2019-10-31 Online:2019-12-31 Published:2020-01-01

PDF

1683

摘要： 实现对驾驶行为语义的精准解析是进一步提升自动驾驶汽车智能化水平的核心技术之一。该 文总结了驾驶行为研究的体系架构；回顾了驾驶行为语义解析发展的 4 个阶段，包括跟驰及换道模型 建立、驾驶模式离线辨识、驾驶模式及意图在线监测、基于无监督学习的语义解析；介绍了4 项驾驶 行为语义解析关键技术（驾驶模式辨识及预测、驾驶行为智能决策、驾驶风格辨识、驾驶行为语义分 割）在国内外的研究现状；介绍了可应用于智能汽车技术研发的语义解析学习算法，包括 Bayes网络、 隐Markov 模型、深度神经网络、强化学习等；讨论了驾驶行为语义解析技术在智能驾驶中驾驶辅助、 智能决策、路径规划等方面的应用。

关键词: 自动驾驶, 智能汽车, 驾驶行为, 语义解析, 学习算法, 拟人化决策, 驾驶辅助, 路径规划

Abstract: It is one of the key technologies in intelligent vehicle development to precisely decode the hidden semantic characteristics from driving signals along time sequences. This paper systematically proposed a framework of semantic analysis in driving behavior research for decision making in autonomous driving. The 4 stages of semantic analysis were summarized including car-following and lane change model, driving maneuver offline recognition, driver intention online prediction, and unsupervised sematic analysis. The state-of-the-art development of the 4 key techniques in semantic analysis was introduced in detail at home and abroad. The key techniques include maneuver recognition and prediction, smart decision making, driving style estimation, and behavior segmentation. The learning algorithms in semantic analysis for autonomous driving development were introduced in detail, including Bayesian network, hidden Markov model, deep neural network, and reinforcement learning. Finally, the applications of semantic analysis in driver assistance, intelligent decision making, and path planning were discussed for better development of autonomous driving technologies.

Key words: autonomous driving, intelligent vehicle, driving behavior, semantic analysis, learning algorithms, human-like decision making, driver assistance, path planning