摘要
为给自动驾驶的路径规划、决策控制等提供精准的空间信息,针对自动驾驶场景下的3D目标检测技术,区分自动驾驶中的典型传感器,综述近年来基于相机、激光雷达、毫米波雷达以及多模态融合下的3D目标检测方法,总结单模态检测的固有缺陷以及多模态融合中存在的不足,得出未来面向自动驾驶的3D目标检测算法可在多模态数据兼容性、多模态数据处理上和时序网络算法进行拓展研究的结论。
In order to provide accurate spatial information for path planning,decision-making and control of autonomous driving,this paper studies the 3 D target detection technology in autonomous driving scene,and distinguishes typical sensors in autonomous driving.It firstly summarizes 3 D target detection methods based on camera,laser radar,millimetre wave radar and multi-modal fusion in recent years.Then,it summarizes the inherent defects of single-mode detection and the shortcomings of multi-modal fusion.Finally,it puts forward the conclusion that the 3 D target detection algorithm for autonomous driving can expand the research on multimodal data compatibility,multimodal data processing and time-series network algorithm.
作者
侯志斌
朱愿
娄静涛
HOU Zhibin;ZHU Yuan;LOU Jingtao(Army Military Transportation University,Tianjin 300161,China)
出处
《军事交通学报》
2022年第8期78-84,共7页
Journal of Military Transportation University
基金
军队学科专业建设项目
关键词
自动驾驶
3D目标检测
多传感器融合
autonomous driving
3D target detection
multi-sensor fusion