基于单目场景的车辆三维目标检测方法综述

打开文本图片集

中图分类号：TP391.4 文献标识码：A 文章编号：2096-4706（2025）09-0016-10

Abstract： In recent years， vehicle 3D Object Detection has attracted much attention in the field of intelligent transportation such as autonomous driving. Compared with 2D detection， 3D detection can accurately estimate the position， size and attitude of the target in 3D space. Because of the advantages of low cost and high data processing efficiency， monocular camera plays a leading role in practical applications. This paper focuses on the 3D Object Detection method in monocular scenes and systematically sorts out its development context. Firstly， according to the source of prior information， the detection methods are divided into three categories based on geometric information， 2D Object Detection and geometric information constraints， and 3D feature estimation. The core ideas， advantages and disadvantages of representative algorithms in each category are analyzed. Secondly， the common data sets and evaluation indexes commonly used in the field are introduced， and the experimental results of typical algorithms are quantitatively compared on the KITTI dataset. Finally， combined with the current research status， the main existing problems in this field are analyzed， and the future development trend is prospected.

Keywords： intelligent transportation; monocular camera; 3D Object Detection; Deep Learning

0 引 言

目前，计算机视觉领域的三大任务包括图像分类、目标检测和图像语义分割。（剩余20213字）