视觉即时定位与建图算法综述  

Survey of visual simultaneous localization and mapping algorithms

作　　者：王华龙 陈彦泽 刘志成 马兴录[1] Wang Hualong;Chen Yanze;Liu Zhicheng;Ma Xinglu(School of Information Science&Technology,Qingdao University of Science&Technology,Qingdao Shandong 266061,China)

机构地区：[1]青岛科技大学信息科学技术学院,山东青岛266061

出　　处：《计算机应用研究》2025年第2期321-333,共13页Application Research of Computers

基　　金：国家自然科学基金项目(62201313);山东省自然科学基金项目(ZR2020QF007)。

摘　　要：视觉即时定位与建图(visual simultaneous localization and mapping,VSLAM)技术利用视觉传感器分析图像信息,使机器人在未知环境中实现自主定位和实时三维地图构建,是机器人导航和自动驾驶等任务的关键。为了给研究人员提供有价值的参考,梳理了VSLAM的研究现状和最新进展。首先,深入探讨了机器人视觉SLAM算法,根据不同的传感器类型,概述了六种主流的视觉SLAM算法。对这些算法的基本原理进行系统分析,并对其中的经典算法进行了精炼总结。进一步地,将视觉SLAM算法分类为基于特征、基于直接法和基于学习的算法三大类,并详细探讨了各自的优缺点。最后,展望了视觉SLAM技术未来的发展方向,重点关注了深度学习、多传感器融合及实时性能优化等关键研究领域。VSLAM technology utilizes visual sensors to analyze image information,enabling robots to achieve autonomous localization and real-time 3D map construction in unknown environments.It is a critical component for tasks such as robot navigation and autonomous driving.To provide valuable references for researchers,this paper reviewed the current state and latest advancements in VSLAM.Firstly,it conducted an in-depth investigation of robot visual SLAM algorithms based on different sensor types,summarizing six major VSLAM algorithms.By systematically analyzing the fundamental principles of these algorithms,it provided concise summaries of classical algorithms.Furthermore,VSLAM algorithms were classified into three main categories:feature-based,direct-based,and learning-based,with detailed discussions of their respective advantages and disadvantages.Finally,it explored future directions for VSLAM technology,focusing on key research areas such as deep lear-ning,multi-sensor fusion,and real-time performance optimization.

关 键 词：视觉SLAM 特征法 直接法 深度学习 未来发展 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置] TP391.41[自动化与计算机技术—控制科学与工程]