单目视觉下基于三维目标检测的车型识别方法综述  

Survey of Vehicle Recognition Methods Based on 3d Object Detection in Monocular Scene

作　　者：王伟[1,2] 唐心瑶 宋焕生 张朝阳[1] WANG Wei;TANG Xin-yao;SONG Huan-sheng;ZHANG Chao-yang(School of Information Engineering,Chang'an University,Xi'an 710064,China;Anhui Science and Technology Information Industry CO.Ltd,Hefei 230088,China)

机构地区：[1]长安大学信息工程学院,西安710064 [2]安徽科力信息产业有限责任公司,合肥230088

出　　处：《小型微型计算机系统》2020年第6期1274-1280,共7页Journal of Chinese Computer Systems

基　　金：教育部联合基金项目(6141A02022610)资助;中央高校基金项目(300102249103)资助;陕西省重点研发计划重点项目(2018ZDXM-GY-047)资助;陕西省社会发展领域项目(2019SF-258)资助.

摘　　要：近年来,车辆三维检测在无人驾驶及智能交通等领域得到了广泛的关注.但当前基于单目视觉的车辆三维检测车型识别方法并没有完善的总结,因此本文对该类方法进行了综述探讨.首先,将基于三维目标检测的车型识别问题分为粗粒度识别和细粒度识别两大类,接着根据不同的类别分别回顾了每类问题的发展历程,重点阐述了每类问题中代表性算法的核心思想及优缺点,然后介绍了两类问题中一些常用的公开数据集并且对它们的特点进行了对比,最后讨论了基于三维目标检测的车型识别目前还存在的一些问题和未来的发展前景.In recent years,vehicle type recognition has received extensive attention in the field of driverless and intelligent transportation.As vehicle recognition methods based on 3 d object detection in monocular scene are not systematic enough,the research status and specific methods of the vehicle recognition problem based on 3 d object detection are summarized and discussed.First of all,the vehicle recognition problem based on 3 d object detection can be divided into two major categories:coarse-grained recognition and finegrained recognition.Then,the development process of each type of problem according to the different categories are reviewed respectively,focusing on the core idea of the algorithms in each type of problems and the advantages and disadvantages of these algorithms.Some of the frequently-used public datasets and their characteristics are introduced.Finally,the existing problems and possible development prospects in the future of the vehicle recognition based on 3 d object detection are discussed.

关 键 词：智能交通 车型识别 粗粒度识别 细粒度识别 三维目标检测 单目相机 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]