计算机科学与探索 ›› 2022, Vol. 16 ›› Issue (12): 2695-2717.DOI: 10.3778/j.issn.1673-9418.2206026

• 综述·探索 • 上一篇    下一篇

激光点云的三维目标检测研究进展

周燕, 蒲磊(), 林良熙, 刘翔宇, 曾凡智, 周月霞   

  1. 佛山科学技术学院 计算机系,广东 佛山 528000
  • 收稿日期:2022-06-06 修回日期:2022-08-31 出版日期:2022-12-01 发布日期:2022-12-16
  • 通讯作者: +E-mail: 2112151112@stu.fosu.edu.cn
  • 作者简介:周燕(1979—),女,江西抚州人,硕士,教授,硕士生导师,CCF会员,主要研究方向为图像处理、计算机视觉、机器学习。
    蒲磊(1998—),男,湖南邵阳人,硕士研究生,主要研究方向为计算机视觉、三维目标检测。
    林良熙(1995—),男,广东湛江人,硕士研究生,主要研究方向为计算机视觉、三维目标检测。
    刘翔宇(1987—),女,河南安阳人,博士,讲师,CCF会员,主要研究方向为计算机视觉、机器学习、人机交互。
    曾凡智(1965—),男,湖北洪湖人,博士,教授,硕士生导师,CCF会员,主要研究方向为计算机视觉、图像处理、数据挖掘。
    周月霞(1978—),女,湖北监利人,硕士,讲师,主要研究方向为信息采集与处理。
  • 基金资助:
    国家自然科学基金(61972091);广东省自然科学基金(2022A1515010101);广东省自然科学基金(2021A1515012639);广东省普通高校重点研究项目(2019KZDXM007);广东省普通高校重点研究项目(2020ZDZX3049);佛山市科技创新项目(2020001003285);广东省教育科学规划课题(2021GXJK445)

Research Progress on 3D Object Detection of LiDAR Point Cloud

ZHOU Yan, PU Lei(), LIN Liangxi, LIU Xiangyu, ZENG Fanzhi, ZHOU Yuexia   

  1. Department of Computer Science, Foshan University, Foshan, Guangdong 528000, China
  • Received:2022-06-06 Revised:2022-08-31 Online:2022-12-01 Published:2022-12-16
  • About author:ZHOU Yan, born in 1979, M.S., professor, M.S. supervisor, member of CCF. Her research inte-rests include image processing, computer vision and machine learning.
    PU Lei, born in 1998, M.S. candidate. His re-search interests include computer vision and 3D object detection.
    LIN Liangxi, born in 1995, M.S. candidate. His research interests include computer vision and 3D object detection.
    LIU Xiangyu, born in 1987, Ph.D., lecturer, member of CCF. Her research interests include computer vision, machine learning and human-computer interaction.
    ZENG Fanzhi, born in 1965, Ph.D., professor, M.S. supervisor, member of CCF. His research interests include computer vision, image proces-sing and data mining.
    ZHOU Yuexia, born in 1978, M.S., lecturer. Her research interest is information acquisition and processing.
  • Supported by:
    National Natural Science Foundation of China(61972091);Natural Science Foundation of Guangdong Province(2022A1515010101);Natural Science Foundation of Guangdong Province(2021A1515012639);Key Research Project of University of Guangdong Province(2019KZDXM007);Key Research Project of University of Guangdong Province(2020ZDZX3049);Science and Technology Innovation Project of Foshan(2020001003285);Educational Science Planning Project of Guangdong Province(2021GXJK445)

摘要:

三维目标检测是近年来新兴的研究方向,其主要任务是对空间中的目标进行定位与识别。目前采用单目或双目视觉的方法来完成三维目标检测任务,其容易受物体遮挡、视点变化和尺度变化的影响,导致检测精度不佳及鲁棒性差等问题。由于激光点云能描述三维场景的信息,在激光点云数据的基础上使用深度学习的方法完成三维目标检测任务,已成为三维视觉领域中研究的热点。针对激光点云的三维目标检测,梳理了近年来相关的研究工作。首先根据输入网络的数据形式,将基于激光点云的三维目标检测方法分为基于原始点云、基于点云投影、基于点云体素化及基于多模态融合的三维目标检测方法,并对各类最具有代表性的方法进行了详细阐述。然后介绍了当前常用的开源数据集及其评价指标,并在数据集上对各类方法进行了性能对比,从多个方面讨论了各类方法的优势及局限性。最后指出当前激光点云的三维目标检测研究存在的不足和难点,并对其未来的发展趋势进行了总结与展望。

关键词: 三维目标检测, 激光点云, 深度学习

Abstract:

3D object detection is a new research direction in recent years, and its main task is the location and recognization of targets in space. The existing methods for 3D object detection using monocular or binocular stereo vision are easily affected by object occlusion, viewpoint changing and scale changing in 3D scene, there will be problems such as poor detection accuracy and robustness. LiDAR point cloud can provide 3D scene information, so using deep learning method to complete 3D object detection based on LiDAR point cloud has become a research hotspot in the field of 3D vision. Aiming at the 3D object detection based on LiDAR point cloud, the relevant research in recent years is reviewed. Firstly, the 3D object detection methods based on LiDAR point cloud are divided into point cloud based, point cloud projection based, point cloud voxelization based and multi-modal fusion based 3D object detection methods according to the data form of network input, and the most representative methods in each category are described in detail. Then common datasets are introduced, and the performance of representative methods is evaluated, and the advantages and limitations of each method are discussed from several aspects. Finally, the shortcomings and difficulties are given, and the future development directions are also discussed and put forward.

Key words: 3D object detection, LiDAR point cloud, deep learning

中图分类号: