[論文レビュー] PL-VINS: Real-Time Monocular Visual-Inertial SLAM with Point and Line Features

Qiang Fu, Jialong Wang+5 more|arXiv (Cornell University)|Sep 16, 2020

Robotics and Sensor-Based Localization参考文献 36被引用数 59

ひとこと要約

PL-VINSは点と線の特徴をリアルタイムのモノクラ VINSに統合し、修正LSDを用いた線検出とPlücker座標ベースの空間線を用いて、同じ更新レートでVINS-Monoよりローカリゼーション精度を向上させます。

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 Plücker 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 Plücker 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.

研究の動機と目的

• challenging environments by incorporating line features.
• Develop a real-time optimization-based VINS that fuses points, lines, and IMU data.
• Replace expensive line processing with a faster line feature extraction and robust representation.
• Demonstrate that line features yield measurable localization gains on public benchmarks.

提案手法

• Modify LSD line segment detector with hidden parameter tuning to accelerate line feature extraction.
• Reject short line features using a length-based pruning strategy to focus on dominant lines.
• Represent space lines using Plücker coordinates and model line reprojection residual as a point-to-line distance for optimization.
• Triangulate points and lines and optimize a sliding-window objective that includes point, line, and IMU residuals.
• Use a four-parameter orthonormal representation for lines to improve convergence in optimization.
• Maintain a loop-closure module similar to VINS-Mono for global consistency.

実験結果

リサーチクエスチョン

• RQ1 Can real-time monocular VINS be significantly improved by integrating line features without sacrificing speed?
• RQ2 Does a faster, tuned line detector coupled with Plücker-based line representation improve localization accuracy over point-only VINS-Mono?
• RQ3 What are the trade-offs between line feature quantity and pose estimation performance in a real-time SLAM system?

主な発見

Dataset (Sequence)	VINS-Mono ATE RMSE w/o loop	PL-VINS ATE RMSE w/o loop	VINS-Mono ATE RMSE w/ loop	PL-VINS ATE RMSE w/ loop
MH-04- difficult	0.375	0.270	0.220	0.202
MH-05- difficult	0.296	0.272	0.242	0.252
V1-02- medium	0.095	0.105	0.084	0.092
V1-03- difficult	0.176	0.156	0.170	0.152
V2-03- difficult	0.293	0.237	0.280	0.182
Mean	0.247	0.208	0.199	0.175

• PL-VINS achieves higher localization accuracy than VINS-Mono on EuRoc sequences at the same pose update frequency (ATE RMSE improvements up to 16% on average without loop, and 12% with loop).
• A modified LSD for line detection runs at least three times faster than the original LSD while maintaining practical accuracy for pose estimation.
• Line features fused in a sliding-window optimization (points, lines, IMU) yield improved RPE performance and yield more best results across evaluated sequences.
• The system can run at 10 Hz on a low-power CPU for monocular VINS with points and lines, outperforming a comparable PL-VIO variant in speed due to efficient line processing.
• Line feature constraints are effective even without loop closure, and the approach rebuilds space lines for better visualization of trajectories.