基于六维语义空间的自动驾驶风险评估研究  

Risk assessment of autonomous vehicle based on six-dimensional semantic space

作　　者：陈亚男 李昂[2] 吴丹 CHEN Yanan;LI Ang;WU Dan(College of Telecommunications&Information Engineering,Nanjing University of Posts and Telecommunications,Nanjing 210003,China;Institute of Space-Air-Ground-Sea Integrated Communication Technology,Nanjing University of Posts and Telecommunications,Nanjing 210003,China;College of Communications Engineering,Army Engineering University,Nanjing 210007,China)

机构地区：[1]南京邮电大学通信与信息工程学院,江苏南京210003 [2]南京邮电大学空天地海通信技术一体化研究院,江苏南京210003 [3]陆军工程大学通信工程学院,江苏南京210007

出　　处：《通信学报》2024年第1期77-93,共17页Journal on Communications

基　　金：国家自然科学基金资助项目(No.62231017,No.62122094);江苏高校优势学科建设工程基金资助项目。

摘　　要：针对自动驾驶中风险要素提取不充分、风险场景评估鲁棒性低等问题,提出一种基于六维语义空间的风险评估框架,包括基于六维语义空间的风险要素提取和基于知识图谱的风险场景评估。前者构建六维语义空间并将RGB和红外数据映射其中,利用模态间的关联提取丰富的数据特征,以获得显在和潜在的风险要素。后者通过语义角色标注和实体融合将风险要素凝练为知识图谱,并联合节点补全和风险等级函数设计知识图谱推理方法,实现准确的风险评估。仿真结果表明,较现有的MSMatch和iSQRT-COV-Net方法,所提方法在准确率、漏/虚警率和处理时间上均有优势。To address the problems of inadequate extraction of risk elements and low robustness of risk scenario assessment in autonomous vehicles,a risk assessment framework based on six-dimensional semantic space was proposed,which included risk element extraction based on six-dimensional semantic space and risk scenario assessment based on knowledge graph.Formerly,the semantic space was constructed with RGB and IR data mapped,and rich features were extracted using inter-modal correlations for explicit and potential risk elements.Subsequently,risk elements were distilled into a knowledge graph by semantic role annotation and entity fusion,and an inference method was designed by combining node completion and risk level function for accurate risk assessment.Simulations show that the proposed method surpasses current MSMatch and iSQRT-COV-Net in accuracy,false/missed alarm rate,and processing time.

关 键 词：六维语义空间 知识图谱 风险要素 风险评估 自动驾驶汽车 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]