基于OBB表示的交通场景多传感数据操纵  

OBB-based Multi-Sensor Data Manipulation of Traffic Scenarios

作　　者：刘钰 张驰 李垚辰 李力[4] 刘跃虎[1,3] 郑南宁 LIU Yu;ZHANG Chi;LI Yaochen;LI Li;LIU Yuehu;ZHENG Nanning(Institute of Artificial Intelligence and Robotics,Xi’an Jiaotong University,Xi’an 710049,China;School of Software Engineering,Xi’an Jiaotong University,Xi’an 710049,China;Shaanxi Key Lab of Digital Technology and Intelligent Systems,Xi’an 710049,China;Department of Automation,Tsinghua University,Beijing 100083,China)

机构地区：[1]西安交通大学人工智能与机器人研究所,西安710049 [2]西安交通大学软件学院,西安710049 [3]陕西省数字技术与智能系统重点实验室,西安710049 [4]清华大学自动化系,北京100083

出　　处：《无人系统技术》2019年第5期17-23,共7页Unmanned Systems Technology

基　　金：国家自然科学基金（61973245）

摘　　要：包含诸如危险驾驶行为的边界交通场景,蕴含着影响无人驾驶环境感知智能算法的不利因素,是无人驾驶离线测试中考察算法正确性和环境适应性的有力测试数据,但往往难以直接采集获取,导致样本数据稀缺。对此,旨在探究生成边界交通场景多传感数据的简化3D操纵空间表示方法,通过对多模态传感数据的操纵,实现边界交通场景数据的生成。为平衡操纵空间的几何连续性与图元复杂度需求,引入方向包围盒表征边界交通场景中的交通参与者与道路。进一步,为降低操纵空间的构建开销,提出局部路面渐进式OBB构建方法,动态增加道路几何约束,改善了视觉传感数据的生成效果。实验结果表明,道路环境感知算法对所生成的边界交通场景数据表现敏感,可以用于丰富无人驾驶现有感知智能算法离线测试的测试用例。Critical traffic scenarios,such as scenarios with dangerous driving behaviors,contain challenging conditions that affect the performance of unmanned vehicles’ cognitive algorithm. They are powerful test data in simulation testing for autonomous vehicles. However,the acquisition of critical traffic scenarios is difficult,resulting in few samples.Faced with this,we aim to explore the simplified representation of 3 D manipulating space for generating multi-sensor data of critical traffic scenarios. In this paper,this data is generated by manipulating the existing multimodal sensor data. In order to balance the requirements of excellent geometric continuity and low graphics complexity,OBB(Oriented Bounding Box)is introduced to represent the traffic participants and roads in the critical traffic scenarios. Furthermore,for low expense of manipulating space’s construction,we propose a gradual construction of bounding box for regional road areas method to add road geometric constraints,which improves the visual effect. The experimental results show that the cognitive algorithm is sensitive to the generated critical traffic scenarios,so they can enrich the testing dataset of simulation testing for autonomous vehicles.

关 键 词：方向包围盒 操纵空间表征 边界交通场景 视觉数据生成 数据操纵 无人驾驶离线测试 

分 类 号：TP37[自动化与计算机技术—计算机系统结构]