动态场景下基于深度学习的MR VSLAM综述  

A Survey of VSLAM for Mobile Robots Based on Deep Learning in Dynamic Scenes

作　　者：高春艳[1] 郭超阳 孙凌宇[1] 张明路[1] GAO Chun-yan;GUO Chao-yang;SUN Ling-yu;ZHANG Ming-lu(School of Mechanical Engineering,Hebei University of Technology,Tianjin 300401,China)

机构地区：[1]河北工业大学机械工程学院,天津300401

出　　处：《计算机仿真》2025年第3期466-471,共6页Computer Simulation

基　　金：国家重点研发计划(2022YFB4701101);国家自然科学基金资助项目(U1913211)。

摘　　要：视觉同时定位与地图构建(VSLAM)在动态场景中性能下降和缺乏鲁棒性已成为其实际应用的主要障碍。为应对复杂场景及其参数高度动态的挑战,将深度学习网络与VSLAM系统相结合,分割出场景中的静态背景和动态目标,实现移动机器人的精准定位。介绍已经用于VSLAM系统的深度学习网络,综合分析动态场景下基于深度学习的移动机器人VSLAM系统的研究进展,从动态目标删除与静态背景重建、动态目标跟踪与重建两方面进行综述,并对其深入应用做出展望。The performance degradation and lack of robustness of the visual simultaneous localization and mapping(VSLAM)system in dynamic scenes have become the main obstacles to its practical application.In order to cope with the challenge of complex scenes and their highly dynamic parameters,the deep learning network is combined with the VSLAM system to separate the static background and dynamic target in the scene to achieve accurate positioning of mobile robots.This paper introduces deep learning networks that have been used in VSLAM systems,and comprehensively analyzes the research progress of VSLAM systems for mobile robots based on deep learning in dynamic scenes.A survey is given from two aspects:dynamic target deletion and static background reconstruction,dynamic target tracking and reconstruction,and making prospects for their deep application.

关 键 词：移动机器人 视觉同时定位与地图构建 深度学习 动态场景 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置] TP391.9[自动化与计算机技术—控制科学与工程]