汽车工程 ›› 2022, Vol. 44 ›› Issue (6): 808-820.doi: 10.19562/j.chinasae.qcgc.2022.06.002

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

基于行人视野注意力场的人车微观交互模型

李文礼1,2(),肖凯文1,石晓辉1,梁锋华2,黎平2   

  1. 1.重庆理工大学,汽车零部件先进制造技术教育部重点实验室,重庆  400054
    2.重庆长安汽车股份有限公司,重庆  400020
  • 收稿日期:2021-12-01 修回日期:2021-12-30 出版日期:2022-06-25 发布日期:2022-06-28
  • 通讯作者: 李文礼 E-mail:liwenli@cqut.edu.cn
  • 基金资助:
    重庆市自然科学基金面上项目(cstc2021jcyj-msxmX0183);重庆市留学人员回国创业创新支持计划资助项目(cx2021070);国家自然科学基金(51805061)

Pedestrian-Vehicle Micro-Interaction Model Based on Attention Field of Pedestrian Vision

Wenli Li1,2(),Kaiwen Xiao1,Xiaohui Shi1,Fenghua Liang2,Ping Li2   

  1. 1.Chongqing University of Technology,Key Laboratory of Advanced Manufacturing Technology for Automobile Parts,Ministry of Education,Chongqing  400054
    2.Chongqing Changan Automobile Co. ,Ltd. ,Chongqing  400020
  • Received:2021-12-01 Revised:2021-12-30 Online:2022-06-25 Published:2022-06-28
  • Contact: Wenli Li E-mail:liwenli@cqut.edu.cn

摘要:

从行人视觉认知角度出发,提出一种基于行人视野注意力场的人车微观交互模型。构建视野注意力场驱动行人视野域,利用人工势场驱动行人运动,利用目标捕捉算法来控制行人视野域对目标的捕捉。为了验证模型的有效性,使用无人机采集鸟瞰视角下的人车交互数据并进行处理分析,将行人过街风格分为保守、谨慎和冒险 3种类型,在Pygame平台下搭建仿真场景和交互模型,把不同行人过街风格的交互数据作为模型输入,以模型输出的行人时空轨迹与采集的真实时空轨迹之间的相似度进行实验对比。结果表明,建立的基于行人视野注意力场的人车微观交互模型比常规人工势场模型准确性提高了25.48%,能够有效地复现实际交通场景中的人车交互过程。

关键词: 人车交互, 行人过街风格, 行人视野注意力场, 人工势场

Abstract:

From the perspective of pedestrian visual cognition, a pedestrian-vehicle micro-interaction model based on the atten-tion field of pedestrian vision is proposed. The attention field of pedestrian vision is constructed to drive the pedes-trian field of vision, and the artificial potential field is used to drive the pedestrian movement,. The target capture algorithm is used to control the target capture in the pedestrian visual field. In order to verify the effectiveness of the model, drones are used to collect the pedestrian-vehicle interaction data from the bird's-eye perspective and analyze them. Pedestrian crossing styles are divided into three types: conservative, cautious and adventurous. Simulation scenarios and interactive models are built on the Pygame platform, then, different types of interactive data are used as the model input, the similarity between the pedestrian spatiotemporal trajectory output by the model and the collected real spatiotemporal trajectory is experimentally compared. The results show that the pedestrian-vehicle mi-cro-interaction model based on pedestrian visual attention field is 25.48% more accurate than the conventional artifi-cial potential field model, and it can effectively reproduce the pedestrian-vehicle interaction process in the actual traf-fic scene.

Key words: pedestrian-vehicle interaction, pedestrian crossing style, pedestrian attention field of vision, artificial potential field