Structured Line Feature and Merge Strategy Localization Algorithm Based on Constraints

• Published in: IEEE access Vol. 12; pp. 99957 - 99967
• Main Authors: Zhang, Xiguang, Wu, Renhan
• Format: Journal Article
• Language: English
• Published: IEEE 2024
• Subjects: Cameras; Data mining; Feature extraction; line feature extraction; PL-VINS; Pose estimation; Real-time systems; redundant line removal; Simultaneous localization and mapping; subpixel corner points; Visualization
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2024.3409946

In scenes characterized by weak textures and high motion speeds, traditional visual inertial odometer (VIO) systems face challenges, including reduced accuracy and inadequate real-time performance. These deficiencies may cause the robot to lose some frames in actual operation and the actual error is too large. To address these degradation issues, we propose a VIO system based on point and line fusion features. The system incorporates novel methods for extracting both point and line features. Firstly, subpixel corner extraction is employed to enhance the accuracy of the point feature extraction algorithm. Secondly, for the line feature extraction algorithm, we utilize the FLD line feature extraction method which significantly improves its speed in most environments and enhances its real-time performance. Additionally, to ensure accurate and stable tracking of line features, we introduce a novel idea-combining approach after optimizing these features that reduces pose estimation errors and enhances overall algorithm precision. In the experiment, we found that in the process of online feature extraction, the same line was repeatedly detected in the previous frame and the next frame, and the detected line would become a new line, which would increase the pose estimation error. Therefore, we proposed an optimization method to delete redundant lines for the triangulated line features. Experimental results demonstrate that our proposed method outperforms PL-VINS in terms of both accuracy and speed of line feature extraction on commonly used EuRoC datasets.