[Paper Review] PL-VINS: Real-Time Monocular Visual-Inertial SLAM with Point and Line

Qiang Fu, Jialong Wang+4 more|arXiv (Cornell University)|Sep 16, 2020

Robotics and Sensor-Based LocalizationEngineering23 references22 citations

TL;DR

PL-VINS proposes a real-time monocular visual-inertial SLAM system that integrates point and line features to enhance localization accuracy. By optimizing line extraction with a modified LSD algorithm and modeling line landmarks via Plücker coordinates with iterative minimization of midpoint-to-line distance, it reduces location error by 12–16% compared to VINS-Mono on a 1.1 GHz CPU without GPU acceleration.

ABSTRACT

Leveraging line features to improve location accuracy of point-based visual-inertial SLAM (VINS) is gaining importance as they provide additional constraint of scene structure regularity, however, real-time performance has not been focused. This paper presents PL-VINS, a real-time optimization-based monocular VINS method with point and line, developed based on state-of-the-art point-based VINS-Mono \cite{vins}. Observe that current works use LSD \cite{lsd} algorithm to extract lines, however, the LSD is designed for scene shape representation instead of specific pose estimation problem, which becomes the bottleneck for the real-time performance due to its expensive cost. In this work, a modified LSD algorithm is presented by studying hidden parameter tuning and length rejection strategy. The modified LSD can run three times at least as fast as the LSD. Further, by representing a line landmark with Pl\{u}cker coordinate, the line reprojection residual is modeled as midpoint-to-line distance then minimized by iteratively updating the minimum four-parameter orthonormal representation of the Pl\{u}cker coordinate. Experiments in public EuRoc benchmark dataset show the location error of our method is down 12-16\% compared to VINS-Mono at the same work frequency on a low-power CPU @1.1 GHz without GPU parallelization. For the benefit of the community, we make public the source code: extit{https://github.com/cnqiangfu/PL-VINS

Motivation & Objective

• To improve localization accuracy in monocular visual-inertial SLAM by incorporating line features as structural constraints.
• To address the real-time performance bottleneck caused by expensive line extraction in existing methods, particularly using LSD.
• To develop an efficient line representation and optimization method using Plücker coordinates for robust line landmark modeling.
• To achieve high accuracy and real-time performance on low-power CPUs without GPU acceleration.
• To release open-source code to support community adoption and further research.

Proposed method

• A modified LSD algorithm is proposed with tuned hidden parameters and a length rejection strategy to accelerate line extraction by at least threefold.
• Line landmarks are represented using Plücker coordinates, enabling a compact four-parameter orthonormal representation.
• The line reprojection residual is modeled as the distance from a 3D point midpoint to a 3D line, minimizing this distance through iterative optimization.
• The optimization process iteratively updates the Plücker coordinate representation to refine line landmark estimation.
• The system integrates point and line features within an optimization-based VINS framework, extending VINS-Mono with hybrid feature tracking.
• The method is designed for real-time operation on low-power CPUs, avoiding GPU dependency.

Experimental results

Research questions

• RQ1 Can integrating line features into monocular VINS improve localization accuracy without sacrificing real-time performance?
• RQ2 How can line extraction be accelerated sufficiently to support real-time SLAM on low-power hardware?
• RQ3 What is the most effective way to represent and optimize line landmarks in a visual-inertial state estimation framework?
• RQ4 How does the use of Plücker coordinates improve line feature representation and reprojection error minimization?
• RQ5 To what extent can line features reduce localization error compared to point-only VINS on standard benchmarks?

Key findings

• The modified LSD algorithm achieves at least three times faster line extraction than the original LSD, enabling real-time performance.
• PL-VINS reduces location error by 12–16% compared to VINS-Mono on the EuRoC benchmark when running at the same frequency on a 1.1 GHz CPU.
• The use of Plücker coordinates with iterative minimization of midpoint-to-line distance improves the accuracy of line landmark representation.
• The system achieves real-time performance without GPU acceleration, making it suitable for low-power embedded platforms.
• The open-source release of the code enables reproducibility and facilitates future research in hybrid point-line SLAM.