基于视频的电动自行车分心事故原因研究  

A study on the severity of E‑bikes distracted accidents based on videos

作　　者：冯战雨 张健[1,2,3] 姜涵[1,2] FENG Zhanyu;ZHANG Jian;JIANG Han(Jiangsu Key Laboratory of Urban ITS,Southeast University,Nanjing 211189;School of Transportation,Southeast University,Nanjing 211189,China;Institute of Technology,Tibet University,Lhasa 850000,China)

机构地区：[1]东南大学江苏省城市智能交通重点实验室,江苏南京211189 [2]东南大学交通学院,江苏南京211189 [3]西藏大学工学院,西藏拉萨850000

出　　处：《现代交通与冶金材料》2023年第5期43-48,共6页Modern Transportation and Metallurgical Materials

基　　金：国家重点研发计划项目(2021YFB1600500);西藏自治区自然科学基金资助项目(XZ202201ZR0040G)。

摘　　要：近年来,中国城市道路交通系统中电动自行车分心事故呈现多发态势。基于2017~2022年的电动自行车事故视频,从事故时间特性、空间特性、交通参与者特性、分心行为、风险行为及事故严重程度六个方面对468起事故进行信息提取。采用随机森林模型、Logistic回归模型、决策树模型和支持向量机模型对数据进行拟合度分析,各模型分析精准度分别为88.0%,77.8%,82.1%和73.5%,表明随机森林模型拟合度效果最优。采用随机森林模型量化分析事故中各类潜在因素对事故严重程度影响的显著性,并进行排序,结合事故频率次数的二维热力图分析了事故因素之间的潜在联系。In recent years,e-bike distraction accidents in China's urban road traffic system have shown a high incidence.Based on the e-bike accident videos from 2017 to 2022,information is extracted from 468 accident videos in six aspects,that is,accident time characteristics,spatial characteristics,traffic participant characteristics,distraction behavior,risk behavior,and accident severity.The random for⁃est model,logistic regression model,decision tree model and support vector machine model are used to analyze the fit of the data,and the analysis accuracies of each model are 88.0%,77.8%,82.1%,and 73.5%,respectively,indicating that the random forest model has the best fit effect.The random forest model is then used to quantify and rank the significance of the influence of various potential fac⁃tors in accidents on accident severity.A two-dimensional heat map combining the number of accident frequencies is used to analyze the potential links between accident factors.

关 键 词：交通安全 电动自行车 事故视频 骑行分心 事故严重程度 

分 类 号：U491.3[交通运输工程—交通运输规划与管理]