自动驾驶汽车测试用车载感知分析系统

针对自动驾驶汽车道路测试过程中缺少高效的环境智能感知与测试数据记录分析设备的问题,依据轻量化、高效化、均衡化原则,设计了一种自动驾驶汽车测试用车载感知分析系统。该系统具备了环境感知分析与数据记录两大功能,在自动驾驶道路测试安全性评估与违法违规行为监管方面发挥重要的作用。本系统感知分析模块采用纯视觉方案,主要用于本车位置与姿态估计、车道线检测、交通参与者检测,涉及了双目定位与地图构建同步(ORB_SLAM2_Stereo)、基于深度学习的三维车道线检测(Gen_LaneNet)及单目三维目标检测(RTM3D)等算法。系统数据记录模块引用ASAM OSI协议来确保数据的兼容性与复用性,同时采用谷歌Protocol Buffer数据交换格式进行数据存储与读取来减轻硬件资源开销,提高系统实时性。实际检测业务的应用效果表明,本文提出的系统能够满足自动驾驶检测机构或示范应用管理单位的测试和监管需求。

In view of the lack of efficient environment intelligent perception and test data recording and analysis equipment in the road test of autonomous vehicles,an in-vehicle perception analysis system for autonomous vehicle test is designed based on the principles of lightweight,high efficiency and equalization.The system has two main functions of environment perception analysis and data recording,which plays an important role in the safety evaluation of autonomous vehicle road test and the supervision of illegal behaviors.The perception analysis module adopts a pure visual scheme,which is mainly used for the position and posture estimation,lane line detection and traffic participant detection,and involves algorithms like the binocular positing and map construction synchronization(ORB_SLAM2_Stereo),the deep learning based 3D lane line detection(Gen_LaneNet)and the monocular 3D object detection(RTM3D).To ensure compatibility and reusability,the data recording module adopts ASAM OSI protocol,and uses the Google Protocol Buffer data exchange format for data storage and reading to reduce hardware resource overhead and improve system real-time performance.The practical application results show that the system proposed in this paper can meet the testing and supervision requirements of the automatic driving testing agency or the demonstration application management organization.