摘要
为提升网联环境下车载信息的传递效率,提出了一种驾驶人信息认知地图构建方法。在驾驶人交通场景信息认知中引入认知地图概念,通过对20名驾驶人进行面对面访谈,获取视距内、外(超空间距离)交通场景语义描述,采用要素频率统计法确定驾驶人信息认知地图要素,设计概念图形完成可视化映射。构建轻量化深度学习MobileNet V2模型,实现驾驶人信息认知地图认知要素标签自动生成,并分别基于SE(Squeeze-and-Excitation Module)、CBAM(Convolutional Block Attention Module)、CA(Coordinate Attention Module)注意力模块对模型进行改进,通过Opencv算法实现可视化认知地图的自动生成。开展20名驾驶人的实车试验,获取12000组交通场景认知集作为测试数据,通过人工标注的方法获取认知要素标签。结果表明:驾驶人信息认知地图具有道路类型、车道数、自身车道、目标类型、方向、距离、危险程度7个认知要素;MobileNet V2、MobileNet V2-SE、MobileNet V2-CBAM和MobileNet V2-CA模型在测试集上的平均准确率分别为89.46%、90.99%、91.29%、91.14%。提出的驾驶人信息认知地图生成方法,能够简化、概括表征危险驾驶场景,有助于提升网联环境下的信息传递效率和安全性。
A method for constructing a driver information cognitive map is proposed to improve the efficiency of vehicle information transmission in connected environments.The concept of a cognitive map was introduced into the drivers'traffic scene information cognition.Face-to-face interviews were conducted with the 20 drivers.The drivers'semantic descriptions of visual internal and over-the-horizon traffic scenes were obtained.Elemental frequency statistics were used to determine cognitive map elements,and cognitive maps were visualized using concept graphics.A MobileNet V2 model was constructed for the automatic generation of cognitive element labels for driver information cognitive maps.The model was improved based on squeeze-and-excitation,convolutional block attention,and coordinate attention modules.Twenty drivers performed real-world vehicle tests and collected 12000 traffic scene sets as test data.The cognitive element labels were manually obtained.The experimental results show that the cognitive map has seven cognitive elements:road type,number of lanes,self-lane,target type,direction,distance,and degree of danger.The average accuracies of MobileNet V2,MobileNet V2-SE,MobileNet V2-CBAM,and MobileNet V2-CA on the test set were 89.46%,90.99%,91.29%,and 91.14%,respectively.The proposed driver information cognitive map generation method can simplify and summarize the representation of dangerous driving scenarios,which is helpful for improving the efficiency and safety of information transmission in an intelligently connected environment.
作者
李靖宇
冯忠祥
张卫华
窦思伟
周正
钱昱昭
LI Jing-yu;FENG Zhong-xiang;ZHANG Wei-hua;DOU Si-wei;ZHOU Zheng;QIAN Yu-zhao(School of Civil and Hydraulic Engineering,Hefei University of Technology,Hefei 230009,Anhui,China;School of Automobile and Traffic Engineering,Hefei University of Technology,Hefei 230009,Anhui,China;School of Information and Computer,Anhui Agricultural University,Hefei 230036,Anhui,China)
出处
《中国公路学报》
EI
CAS
CSCD
北大核心
2023年第9期302-314,共13页
China Journal of Highway and Transport
基金
国家自然科学基金项目(52272345,71971073,52372326)
安徽省重点研究与开发计划项目(2022k07020005,202304a05020050)
安徽省住房城乡建设科学技术计划项目(2021-YF43,2023-YF095)
安徽省研究生学术创新项目(2022xscx020)。
关键词
交通工程
认知地图
深度学习
智能网联
车载信息
交通语义
traffic engineering
cognitive map
deep learning
intelligent network
in-vehicle infotainment
traffic scene semantic
