摘要
典型场景的边界需要从实际道路中发生的原始场景数据中提取,但人工场景标注方法效率低、成本高.因此,研究真实道路中自动驾驶场景的获取对自动驾驶汽车的测试具有重要意义,而切入场景是典型的危险场景.本研究基于一辆搭载L2级自动驾驶系统的车辆,改装多源异构传感器的场景采集系统,比较了类驾驶员与自动驾驶系统在速度保持和车道居中行驶的能力.提出了基于预碰撞区域的切入场景自动识别算法,并基于SCANeR软件构建了识别算法评估测试平台,设计了大量测试用例生成程序对算法进行仿真测试.在试验场进行了11组实车切入场景测试,对识别算法进行了评价,并将仿真结果与实际试验结果进行了比较.
The boundaries of a typical scene need to be extracted from the data of original scene that occurs on the actual road,but the manual scene annotation method is inefficient and costly.Therefore,the study of automatic driving scene acquisition in real road is of great significance to the test of autopilot vehicle.The cut-in scene is a typical dangerous scene.Based on a vehicle equipped with L2 automatic driving system,the scene acquisition system of multi-source heterogeneous sensors was modified.The ability of driver and autopilot system to maintain the speed and lane was compared.Then,an automatic cut-in scene recognition algorithm based on pre-collision area was proposed.A recognition algorithm evaluation test platform was constructed based on SCANeR software,and a large number of test case generation programs were designed to simulate and test the algorithm.Then,11 groups of real vehicle cut-in scene tests were carried out in the test field,the recognition algorithm is evaluated,and the simulation results were compared with the actual test results.
作者
曾杰
廖伟
丁雪聪
胡雄
ZENG Jie;LIAO Wei;DING Xuecong;HU Xiong(National Intelligent Network Automobile Quality Inspection and Testing Center(Chongqing)/China Merchants Testing Vehicle Technology Research Institute Co.,Ltd.,Chongqing 401112,China;College of Mechanical and Vehicle Engineering,Chongqing University,Chongqing 400044,China)
出处
《西南大学学报（自然科学版）》
CAS
CSCD
北大核心
2022年第3期20-28,共9页
Journal of Southwest University(Natural Science Edition)
基金
重庆市自然科学基金面上项目(cstc2020jcyj-msxmX0496)
川渝联合实施重点研发项目(cstc2021jscx-cylh0026)
重庆市市场监督管理局科技项目(CQSJKJ2021010)。
关键词
自动驾驶
场景
识别
验证
autonomous vehicle
scenario
detection
verification
