自动驾驶的单目视觉目标检测与分割技术综述

Development of object detection and segmentation based on monocular vision of autonomous driving

环境感知是自动驾驶的核心功能,主要包括目标检测和语义分割技术。近年随着深度学习的发展涌现大量优秀的环境感知算法,已有相关文献对部分功能算法进行总结,但还缺少对单目视觉下环境感知技术整体性的概述,基于此本文总结了近年单目相机感知技术的发展,包括2D,3D目标检测和语义分割技术。介绍了自动驾驶环境感知技术概要,围绕深度学习对目前不同任务下主流的算法进行分类和比较,提出感知技术发展可能遇到的挑战,并给出一些有前景的研究方向。

Environment perception is a core function of autonomous driving,mainly including target de⁃tection and semantic segmentation techniques.In recent years,a large number of excellent environ⁃ment perception algorithms have emerged with the development of deep learning,and relevant litera⁃ture has summarized some of the functional algorithms.However,a holistic overview of environment perception technology under monocular vision is still lacking.Based on this,this paper summarizes the development of monocular camera perception technology in recent years,mainly including 2D and 3D object detection and semantic segmentation techniques.An overview of environment perception tech⁃niques for autonomous driving is presented.The mainstream algorithms for different tasks are then clas⁃sified and compared mainly around deep learning methods.The challenges that may be encountered in the development of perception techniques are presented,and some promising research directions are given.