PKL-VINS a robust monocular visual-inertial SLAM with point and keypoint-line features
Lists
Only you see this
Abstract
Abstract Robust visual-inertial simultaneous localization and mapping systems are fundamental for environmental perception in outdoor navigation scenarios, including autonomous driving and field surveying. However, traditional point-feature-based methods are prone to feature degradation under challenging conditions such as drastic illumination changes and low-texture regions. Furthermore, existing line-feature-based methods often suffer from limitations like low efficiency and high error rates in cross-frame line segment matching. To overcome these limitations, this paper proposes a multi-level line segment optimization framework that integrates line segment quality enhancement and optimized cross-frame trajectory matching strategies. First, we design an adaptive line segment filter based on environmental scale assessment, utilizing dynamic length thresholds to construct a refined set of line segments. Second, we enhance the line segment representation capability of the EDLines algorithm by incorporating a dynamic extension strategy and keypoint detection. Third, we propose a line segment matching algorithm integrating dynamic time warping, which improves matching robustness while maintaining computational efficiency. Finally, we formulate a feature-fusion state estimation model by tightly coupling the optimization of line feature reprojection errors with point features. Experimental results on the EuRoC and NTU VIRAL datasets demonstrate that the proposed method significantly outperforms state-of-the-art approaches in line segment detection, localization accuracy, and matching time.
BibTeX
@inproceedings{chen2025pkl,
author = {Zailiang Chen and Zeping Liu and Hailan Shen and Shuo Huang and Yuehong Tong},
title = {{PKL-VINS a robust monocular visual-inertial SLAM with point and keypoint-line features}},
booktitle = {Measurement},
year = {2025},
doi = {10.1088/1361-6501/adeeb0},
}Loading related content...