自动驾驶感知系统安全性定量测评方法研究  

Research on Safety Evaluation Method of Autonomous Vehicles’Perception Systems Based on Quantitative Index

作　　者：宋国正 朱可涵 李宏强 孟祥坤 李钟琦[1] Song Guozheng;Zhu Kehan;Li Hongqiang;Meng Xiangkun;Li Zhongqi(Instrumentation Technology and Economy Institute,Beijing 100055;Navigation College of Dalian Maritime University,Dalian Liaoning 116026)

机构地区：[1]机械工业仪器仪表综合技术经济研究所,北京100055 [2]大连海事大学航海学院,辽宁大连116026

出　　处：《中国仪器仪表》2025年第2期17-24,共8页China Instrumentation

基　　金：“国家重点研发计划”(课题编号:2021YFC0600205)资助。

摘　　要：自动驾驶感知系统感知环境和位姿信息,其表现直接影响运行风险。本文提出基于量化指标的自动驾驶感知安全性测评方法,评价在封闭驾驶场景中感知系统的安全性。该方法给出了感知系统风险的定义,然后建立故障树模型,对驾驶场景中感知问题进行系统性分析,提出了不同场景感知风险测评指标,最后基于多场景风险分析结果综合评价感知系统安全性等级。本文结合案例展示了感知安全性定量测评方法的应用。该方法基于量化指标评估感知系统的安全性,可为自动驾驶感知安全性保障、感知系统升级优化以及决策系统的开发提供数据支撑。Perception systems of autonomous vehicles collect and process environmental and vehicle pose information,whose performance directly affect operational risks.This paper proposes a safety evaluation method of perception systems in closed driving scenarios based on quantitative indicators.The method defines perceived system risks.Then,perception problems in driving scenes are analyzed using the fault tree models,and the perceived risk assessment indicators are proposed.Finally,based on the results of multi-scene risk analysis,the safety of perception systems is comprehensively evaluated.The application of the proposed method is demonstrated using a case study.This method can determine the safety of perception systems based on quantitative indicators.It can provide data support for the safety guarantee of autonomous driving perception,the upgrading and optimization of the perception systems,and the development of decision-making systems.

关 键 词：自动驾驶 感知系统 风险评估 故障树 安全性测评 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]