LRPL-VIO: A Lightweight and Robust Visual-Inertial Odometry with Point and Line Features

Area of Science:

• Robotics
• Computer Vision
• Sensor Fusion

Background:

• Visual-inertial odometry (VIO) enhances navigation by combining camera and inertial sensor data.
• Traditional VIO algorithms struggle with performance degradation and increased computational cost in challenging environments.
• Fusing diverse features like points and lines can improve VIO accuracy but often leads to higher processing times.

Purpose of the Study:

• To develop a lightweight and efficient point-line visual-inertial odometry algorithm.
• To address the trade-off between performance and computational cost in VIO systems.
• To enhance robustness and speed in challenging visual-inertial odometry scenarios.

Main Methods:

• Proposed a novel lightweight point-line visual-inertial odometry algorithm (LRPL-VIO).
• Developed a fast line matching method leveraging photometric invariance of feature points between frames.
• Implemented an efficient filter-based state estimation framework for fusing point, line, and inertial data.
• Introduced a unique feature selection scheme to prioritize high-quality line features for state estimation.

Main Results:

• LRPL-VIO significantly reduces front-end processing time through its fast line matching.
• The algorithm demonstrates improved efficiency by selecting only high-quality line features for state estimation.
• Experimental validation on public datasets and real-world tests confirms superior speed and robustness compared to state-of-the-art methods.
• The proposed method excels in challenging scenes where traditional VIO algorithms falter.

Conclusions:

• LRPL-VIO offers a computationally efficient and robust solution for visual-inertial odometry.
• The integration of point and line features, coupled with an efficient state estimation framework, enhances VIO performance.
• This lightweight algorithm presents a promising advancement for real-time navigation applications in complex environments.