融合点线特征的视觉惯性同时定位及建图
Visual-inertia simultaneous localization and mapping based on point-and-line features
姚建均 1李英朝 1吴杨 2唐瑞卓 1于新达 1闫宇坤1
作者信息
- 1. 哈尔滨工程大学 机电工程学院,黑龙江 哈尔滨 150001
- 2. 维沃移动通信有限公司,广东 东莞 523850
- 折叠
摘要
为了解决移动机器人在低纹理场景中的定位精度较低且容易跟踪丢失问题,本文设计了一种点线特征提取和匹配策略,并以此构建了视觉惯性同时定位和建图系统.提出线特征提取和匹配算法,通过改良线特征提取算法的隐藏参数,提高了线特征提取的精度.利用点线特征不同匹配筛选框架减少误匹配的数目,得出了可以应用于视觉惯性同时定位和建图系统的线特征提取匹配算法.在现有视觉惯性框架中引入提出的线特征约束,搭建了能在未知低纹理环境下鲁棒运行的视觉惯性同时定位及建图系统.研究表明:在真实环境中的移动机器人定位实验证明了该系统的精度和鲁棒性优于现有的视觉惯性框架,其室内定位精度提高了 24.2%,走廊定位精度提高了 8%,对于移动机器人在低纹理场景下的高精度定位具有较高价值.
Abstract
The accuracy of mobile robot localization in low-texture scenes is often compromised,leading to frequent tracking loss.To solve this problem,this study proposes an innovative point-line feature extraction and matching strategy incorporated into the visual-inertial simultaneous localization and mapping(SLAM)system.The approach begins by proposing a line feature extraction and matching algorithm.Refining the hidden parameters of the line feature extraction algorithm improves accuracy.Subsequently,diverse matching screening frameworks for point-line features are employed to reduce mismatches.This approach results in a line feature extraction matching algorithm suitable for the visual-inertial SLAM system.By integrating the proposed line feature constraint into the current vis-ual-inertial framework,this study establishes a robust visual-inertial SLAM system suitable for operation in unknown low-texture environments.Experimental validation with a mobile robot in a real-world setting demonstrates superior accuracy and robustness of the proposed strategy compared with those of the existing visual-inertial framework.The system enhances indoor localization accuracy by 24.2%and corridor localization accuracy by 8%,providing sub-stantial value for high-precision mobile robot localization in low-texture scenes.
关键词
同时定位及建图/点线特征/视觉惯性里程计/单目视觉/移动机器人感知/特征匹配/低纹理/回环检测Key words
simultaneous localization and mapping/point-and-line features/visual-inertial odometer/monocular vi-sion/mobile robot perception/feature matching/low texture/loop detection引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
维普
万方数据