基于眼动数据分析交通环境要素对驾驶员视觉负荷的影响  

Analyzing the Impact of Traffic Environment Elements on Driver Visual Load Based on Eye Movement Data

作　　者：白玉[1] 冷帅 BAI Yu;LENG Shuai(School of Transportation Engineering,Tongji University,Shanghai 201804,China)

机构地区：[1]同济大学交通运输工程学院,上海201804

出　　处：《交通工程》2024年第5期120-128,共9页Journal of Transportation Engineering

基　　金：新型混合交通流背景下交通适驾性影响机理及改善对策研究;批准号:52272320。

摘　　要：研究旨在通过提出1种基于眼动数据得到的负荷评价指标—眼动离散程度,确定不同视觉负荷下眼动离散程度的划分阈值,并利用该指标探究交通环境要素对人工驾驶员视觉负荷影响的程度排序,进而为交通设计提供参考。研究方法为分析DADA2000数据集,利用语义分割算法与眼动特征识别算法,建立眼动离散程度变化机制模型,并将事故前3s眼动点集中区域与语义分割得到的要素进行匹配,进而得到不同视觉负荷下眼动离散程度的划分阈值和交通环境要素对驾驶员视觉负荷的影响程度排序。研究结果发现,在不同道路条件下,驾驶员的眼动离散程度和视觉负荷水平会有所变化,但事故前较短时间内眼动离散程度会降低至原来的1/3,眼动点也会集中在碰撞物或事故主要责任要素上,可通过机器学习算法对各交通环境要素影响的重要程度进行排序。分析研究结果可得到结论:利用眼动离散程度区分驾驶员视觉负荷是否过大的方法具有可行性,并可进一步用于事故防范与预测;驾驶员视觉负荷主要来源的前3位要素是自行车参与者、路侧标志杆、路面状况,影响重要度分别达到0.395、0.137、0.124。The research aims to propose a load evaluation index based on eye movement data-the degree of eye movement dispersion,determine the threshold for dividing the degree of eye movement dispersion under different visual loads,and use this index to explore the degree of influence of traffic environment factors on the visual load of artificial drivers,thereby providing reference for traffic design.The research method is to analyze the DADA2000 dataset,use semantic segmentation algorithms and eye movement feature recognition algorithms,establish a mechanism model for the variation of eye movement dispersion degree,and match the concentrated area of eye movement points in the first 3 seconds before the accident with the elements obtained from semantic segmentation.Then,the threshold for dividing eye movement dispersion degree under different visual loads and the ranking of the impact of traffic environment factors on driver visual load are obtained.The research results found that under different road conditions,the degree of eye movement dispersion and visual load level of drivers will vary,but in a short period of time before the accident,the degree of eye movement dispersion will decrease to one-third of the original,and the eye movement points will also be concentrated on the collision object or the main responsible elements of the accident.Machine learning algorithms can be used to rank the importance of the impact of various traffic environment elements.The analysis of research results can lead to the conclusion that the method of using eye movement dispersion to distinguish whether the driver s visual load is too large is feasible and can be further used for accident prevention and prediction;The top three main sources of driver visual load are bicycle participants,roadside markers,and road conditions,with importance levels of 0.395,0.137,and 0.124,respectively.

关 键 词：交通运输工程 视觉负荷 语义分割 眼动特征识别 眼动离散程度 智能交通系统 交通设计 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术] U491[自动化与计算机技术—计算机科学与技术]