| 引用本文: | 王登贵,傅卫平,周劲草,高志强,宋清源.自动驾驶汽车视野遮挡场景潜在风险评估[J].控制理论与应用,2023,40(6):1023~1033.[点击复制] |
| WANG Deng-gui,FU Wei-ping,ZHOU Jin-cao,GAO Zhi-qiang,SONG Qing-yuan.Potential risk assessment of vision-occluded scenarios for autonomous vehicles[J].Control Theory and Technology,2023,40(6):1023~1033.[点击复制] |
|
| 自动驾驶汽车视野遮挡场景潜在风险评估 |
| Potential risk assessment of vision-occluded scenarios for autonomous vehicles |
| 摘要点击 2697 全文点击 500 投稿时间：2022-03-12 修订日期：2023-06-16 |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| DOI编号 10.7641/CTA.2022.20178 |
| 2023,40(6):1023-1033 |
| 中文关键词 自动驾驶汽车 知识图谱 贝叶斯网络 逻辑推理 潜在风险评估 |
| 英文关键词 autonomous vehicles knowledge graph Rayesian network logical reasoning potential risk assessment |
| 基金项目 陕西省自然科学基金项目(2022JQ–546), 国家自然科学基金青年科学基金项目(52005401) |
|
| 中文摘要 |
| 视野遮挡区域的潜在风险对自动驾驶汽车的行驶安全提出了极大挑战. 面对难以准确有效地预测、评估潜 在风险这一困难, 本文提出了一种基于“知识图谱+逻辑推理+贝叶斯推理”的潜在风险评估方法, 建立了潜在风险预 判模型和潜在风险概率评估模型. 潜在风险预判模型通过构建“城市道路驾驶场景知识图谱”描述了场景中实体与 实体间的交互关系, 并对场景描述信息进行语义转换, 采用SWI-Prolog推理机推断当前场景是否存在潜在风险; 潜 在风险概率评估模型基于贝叶斯网络实现对潜在风险的概率评估. 通过现场实验验证, 所提方法对潜在风险的推 理效果与人类驾驶员推理效果相似, 甚至可弥补人类未及时推理出潜在风险的过失. 方法适用于城市复杂道路交 通, 风险评估结果能为自动驾驶或辅助驾驶的行为决策提供有效依据, 具有潜在应用价值. |
| 英文摘要 |
| The autonomous vehicles safety is challenged by the potential risks arising from the visually occluded areas. A new method of potential risks assessment is put forward based on “Knowledge Graph + Logical Reasoning + Bayesian Reasoning” in the face of difficult to accurately and effectively predict and assess such potential risks, a potential risk prediction model and a potential risk probability inference model are constructed in the paper. The potential risk prediction model describes the interaction between driving entities by constructing the knowledge graph of urban scenarios for autonomous driving, and then infers whether there are potential risks during driving with SWI-Prolog inference engine after the description is semantic transformed; The potential risk probability reasoning model can quantify potential risks by inferring the probability of such risks replying on Bayesian Network. The field experiment prove that inference and assessment of potential risks conducted by the proposed method is similar to that conducted by human drivers, and can even compensate for human driver’s unawareness of potential risks. The method is applicable to complex urban traffic. The results of risk assessment have potential application value as they can provide effective basis for the behavioral decision-making of the automatic driving system. |
|
|
|
|
|