PL-VINS: Real-Time Monocular Visual-Inertial SLAM with Point and Line Features (2009.07462v3)

Abstract: Leveraging line features to improve localization accuracy of point-based visual-inertial SLAM (VINS) is gaining interest as they provide additional constraints on scene structure. However, real-time performance when incorporating line features in VINS has not been addressed. This paper presents PL-VINS, a real-time optimization-based monocular VINS method with point and line features, developed based on the state-of-the-art point-based VINS-Mono \cite{vins}. We observe that current works use the LSD \cite{lsd} algorithm to extract line features; however, LSD is designed for scene shape representation instead of the pose estimation problem, which becomes the bottleneck for the real-time performance due to its high computational cost. In this paper, a modified LSD algorithm is presented by studying a hidden parameter tuning and length rejection strategy. The modified LSD can run at least three times as fast as LSD. Further, by representing space lines with the Pl\"{u}cker coordinates, the residual error in line estimation is modeled in terms of the point-to-line distance, which is then minimized by iteratively updating the minimum four-parameter orthonormal representation of the Pl\"{u}cker coordinates. Experiments in a public benchmark dataset show that the localization error of our method is 12-16\% less than that of VINS-Mono at the same pose update frequency. %For the benefit of the community, The source code of our method is available at: https://github.com/cnqiangfu/PL-VINS.

• The paper introduces a real-time monocular SLAM method that integrates both point and line features for improved localization accuracy.
• It optimizes the LSD algorithm with parameter tuning and employs Plücker coordinates for robust 3D line representation, reducing computation time threefold.
• Evaluated on the EuRoc dataset, the approach achieves a 12-16% reduction in localization error while maintaining real-time performance at 10 Hz.

Overview of PL-VINS: Real-Time Monocular Visual-Inertial SLAM with Point and Line Features

The paper presents PL-VINS, a real-time optimization-based monocular Visual-Inertial SLAM (VINS) method that leverages both point and line features to enhance localization accuracy. This approach builds upon the existing VINS-Mono framework and addresses real-time performance challenges associated with incorporating line features.

Key Contributions

PL-VINS introduces several innovations to improve efficiency without sacrificing accuracy:

1. Modified LSD Algorithm: The authors optimize the existing Line Segment Detector (LSD) algorithm, originally designed for scene structure representation, for the specific task of pose estimation. This modification involves parameter tuning and a length rejection strategy, resulting in a threefold reduction in computation time.
2. Integration of Line Features: By employing line features alongside traditional point features, PL-VINS provides additional scene structure constraints, crucial for environments with regular geometric patterns, such as indoor human-made settings.
3. Space Line Representation: Utilization of Plücker coordinates for 3D line representation enables robust space line estimation, accommodating for variations in viewpoint and occlusion.
4. Optimized Residual Error Modeling: The paper proposes a novel method for minimizing line estimation errors through iterative updates of Plücker coordinates, thereby enhancing the system's accuracy in challenging environments.

Evaluation and Results

The performance of PL-VINS is evaluated against the benchmark dataset EuRoc, demonstrating a reduction in localization error by 12-16% compared to the state-of-the-art VINS-Mono. This is achieved without compromising the system's real-time capabilities, as PL-VINS operates efficiently at 10 Hz, maintaining parity with VINS-Mono. The algorithm effectively balances computational demands with the enhanced accuracies brought by the inclusion of line features.

Implications and Future Directions

The integration of line features in PL-VINS represents a significant step towards more reliable SLAM solutions in complex, man-made environments. Its real-time performance ensures applicability in various domains such as robotics and augmented reality.

Future work may explore extending this method to stereo vision systems or integrating other geometric features like planes for further robustness. Additionally, more sophisticated loop-closure techniques integrating line features could be implemented to minimize drift over time.

Conclusion

PL-VINS effectively combines point and line features, demonstrating improved localization accuracy in real-time monocular SLAM applications. Its innovative approach to line feature extraction and integration provides a foundation for future research aiming to refine SLAM technologies in structured environments.