[Paper Review] Avoiding Degeneracy for Monocular Visual SLAM with Point and Line Features

Hyunjun Lim, Yeeun Kim+3 more|arXiv (Cornell University)|Mar 2, 2021

Robotics and Sensor-Based LocalizationEngineering42 references38 citations

TL;DR

This paper proposes a monocular visual SLAM system that combines point and line features to improve robustness and accuracy, especially in low-texture or varying-light environments. It introduces a degeneracy detection method and a novel structural constraint to correct erroneous 3D line reconstructions caused by epipolar geometry, significantly enhancing line tracking and mapping accuracy, as validated on the EuRoC dataset with 15% lower translation RMSE than baseline methods.

ABSTRACT

In this paper, a degeneracy avoidance method for a point and line based visual SLAM algorithm is proposed. Visual SLAM predominantly uses point features. However, point features lack robustness in low texture and illuminance variant environments. Therefore, line features are used to compensate the weaknesses of point features. In addition, point features are poor in representing discernable features for the naked eye, meaning mapped point features cannot be recognized. To overcome the limitations above, line features were actively employed in previous studies. However, since degeneracy arises in the process of using line features, this paper attempts to solve this problem. First, a simple method to identify degenerate lines is presented. In addition, a novel structural constraint is proposed to avoid the degeneracy problem. At last, a point and line based monocular SLAM system using a robust optical-flow based lien tracking method is implemented. The results are verified using experiments with the EuRoC dataset and compared with other state-of-the-art algorithms. It is proven that our method yields more accurate localization as well as mapping results.

Motivation & Objective

• Address the poor robustness of point features in low-texture or high-illumination-variance environments.
• Overcome the limitation of sparse point-based mapping by incorporating line features for better structural discernibility.
• Solve the degeneracy problem in line-based SLAM, where lines near the epipole lead to ambiguous 3D reconstruction.
• Improve line tracking and 3D reconstruction accuracy through a novel structural constraint and optical-flow-based matching.
• Achieve more accurate and visually interpretable mapping in real-world monocular SLAM scenarios.

Proposed method

• Proposes a simple degeneracy detection method for line features in pure translation motion by analyzing epipolar geometry and line proximity to the epipole.
• Introduces a novel structural constraint that enforces geometric consistency (e.g., parallelism) in 3D line reconstruction to resolve ambiguity during triangulation.
• Employs an optical-flow-based line tracking method that fuses predictions from motion models with detected lines to improve matching robustness.
• Uses Plücker coordinates and orthonormal representations for 3D line feature parameterization and bundle adjustment.
• Applies a sliding window optimization framework with constraints on line feature consistency and reprojection error.
• Integrates point and line features in a unified monocular SLAM pipeline with adaptive feature management.

Experimental results

Research questions

• RQ1 How can degenerate line features—caused by epipolar alignment—be detected in monocular SLAM systems?
• RQ2 What geometric constraints can be applied to correct ambiguous 3D line reconstructions caused by degeneracy?
• RQ3 Can optical-flow-based line tracking improve feature persistence and reduce tracking failure in low-texture environments?
• RQ4 To what extent does incorporating structural constraints improve localization and mapping accuracy in line-based monocular SLAM?
• RQ5 How does the proposed method compare to state-of-the-art algorithms like PL-VINS and VINS-Mono in terms of robustness and accuracy?

Key findings

• The proposed method achieved a 15% reduction in translation RMSE compared to the base VINS-Mono algorithm on the EuRoC dataset.
• Line tracking performance improved by approximately 49% compared to the LSD+LBD method, with a 20% increase in computation time.
• In 7 out of 11 EuRoC sequences, the proposed method showed a 25% smaller error than PL-VINS, demonstrating superior performance in handling degeneracy.
• The structural constraint successfully corrected degenerate lines into plausible geometric configurations, such as parallel lines, improving map quality and visual interpretability.
• The system maintained a state estimation rate of ~20Hz, which is acceptable despite the added computational cost of line feature processing.
• The method effectively mitigates the degeneracy problem in line-based SLAM, enabling more accurate and robust 3D reconstruction in challenging environments.