基于激光雷达的3D目标检测研究综述

A Review on LiDAR-Based 3D Target Detection Research

近年来,随着自动驾驶技术的快速发展,智能汽车对于环境感知技术的需求也越来越高,由于激光雷达数据具有较高的精度,能够更好的获取环境中的三维信息,已经成为了3D目标检测领域研究的热点。为了给智能汽车提供更加准确的环境信息,对激光雷达3D目标检测领域主要研究内容进行综述。首先,分析了自动驾驶车辆各种环境感知传感器的优缺点;其次,根据3D目标检测算法中数据处理方式的不同,综述了基于点云的检测算法和图像与点云融合的检测算法;然后,梳理了主流自动驾驶数据集及其3D目标检测评估方法;最后对当前点云3D目标检测算法进行总结和展望,结果表明当前研究中2D视图法和多模态融合法对自动驾驶技术发展的重要性。

In recent years,with the rapid development of autonomous driving technologies,the demand of intelligent vehicles for environment perception technology is also higher and higher.Due to the high accuracy of LiDAR data that can better obtain the 3D information in the environment,it has become a research hotspot in the field of 3D target detection.In order to provide more accurate environmental information for intelligent vehicles,the main research contents in the field of 3D target detection by LiDAR are summarized.Firstly,the advantages and disadvantages of various environment sensing sensors for self-driving vehicles are analyzed;secondly,according to the different data processing methods in 3D target detection algorithms,the detection algorithms based on point cloud and the detection algorithms fused with image and point cloud are reviewed;then,the mainstream self-driving datasets and their evaluation methods for 3D target detection are sorted out;and finally,the current 3D target detection algorithms for point cloud are summarized and outlooked.The results show the importance of the 2D view method and the multimodal fusion method in the current research for the development of autonomous driving technologies.