In context
Visual SLAM is a core technology for autonomous navigation in mobile robots, AGVs, and AR/VR systems. In 2025, researchers continued to address the challenge of maintaining accurate localization in weak-texture environments, where point-feature-based methods often fail. This work from Anhui University of Technology, published in the journal Robot, targets that gap by integrating line features and refining keyframe selection.
What was reported
The authors proposed PLKF-SLAM, an improved binocular visual SLAM method built on ORB-SLAM3. It adds a line-feature extraction module to compensate for scarce point features in weak-texture scenes. The line reprojection error model fuses weighted distance and angle error functions to reject outliers, and an adaptive factor balances the contribution of line features during bundle adjustment based on the number of point features available.
For keyframe selection, a dynamic threshold strategy based on geometric and photogrammetric principles adjusts to camera motion and frame rate, reducing redundant keyframes and computational load. The method was validated on the EuRoC and UMA-VI datasets. Compared with ORB-SLAM3, average positioning accuracy improved by 52.64% on EuRoC and 63.20% on UMA-VI. In real-world tests, the positioning error was 0.05 m.
“The adaptive factor enables the system to dynamically balance the contributions of point and line features based on the feature distribution of the current scene.”
Why it mattered
This work demonstrated a practical approach to improving visual SLAM robustness in low-texture industrial environments, such as warehouses and corridors, where point features are sparse. The adaptive fusion and dynamic keyframe selection offer a path to more reliable and efficient localization for mobile robots and AGVs, potentially reducing computational costs while increasing accuracy.
Source: 《机器人》期刊 (robot.sia.cn) · Published 2025-03-15 · “基于改进的点线融合和关键帧选择的视觉SLAM方法”