Mobile robot localization method based on point-line feature visual-inertial SLAM algorithm

• Published in: Industrial robot Vol. 52; no. 3; pp. 391 - 402
• Main Authors: Xu, Jintao, Fang, Yu, Gao, Weiwei, Liu, Xintian, Shi, Juanjuan, Yang, Hao
• Format: Journal Article
• Language: English
• Published: Bedford Emerald Publishing Limited 30.04.2025; Emerald Group Publishing Limited
• Subjects: Accuracy; Algorithms; Feature extraction; Inertial platforms; Kernel functions; Light; Localization; Localization method; Optimization; Outliers (statistics); Robot learning; Robots; Simultaneous localization and mapping; Texture; Line features; SLAM; Mobile robot localization; Visual-inertial; Weakly textured environment
• ISSN: 0143-991X; 1758-5791; 1758-5791
• DOI: 10.1108/IR-08-2024-0381

Purpose The purpose of this study is to address the low localization accuracy and frequent tracking failures of traditional visual SLAM methods in low-light and weak-texture situations, and we propose a mobile robot visual-inertial localization method based on the improved point-line features VINS-mono algorithm. Design/methodology/approach First, the line feature information is introduced into VINS-mono. Subsequently, the EDlines line feature extraction algorithm is optimized with a short line merging strategy and a dynamic length suppression strategy to reduce redundant short lines and fragmented segments. In the back-end sliding window optimization, line feature reprojection errors are incorporated, and Huber kernel functions are added to the inertial measurement unit residuals, point-line feature residuals and loop closure constraints to reduce the impact of outliers on the optimization results. Findings Comparison and verification experiments are carried out on the EuRoC MAV Data set and real weakly textured environment. In the real low-light and weak-texture scenarios, the improved mobile robot localization system achieves over 40% higher accuracy compared to VINS-mono. Originality/value The main contribution of this study is to propose a new visual-inertial SLAM method combining point-line features, which can achieve good localization effect in low-light and weak-texture scenes, with higher accuracy and robustness.