[Paper Review] Tightly Coupled Optimization-based GPS-Visual-Inertial Odometry with Online Calibration and Initialization

Shihao Han, Feiyang Deng+2 more|arXiv (Cornell University)|Mar 5, 2022

Robotics and Sensor-Based Localization8 citations

TL;DR

This paper presents a tightly coupled optimization-based GPS-visual-inertial odometry system that jointly minimizes visual reprojection, IMU preintegration, and a novel GPS residual to reduce trajectory drift in long-term operations. It introduces online calibration for GPS-IMU extrinsic and time offset, along with a fast initialization method, achieving superior performance on EuRoC and KAIST datasets, including robustness to visual or GPS outages.

ABSTRACT

In this paper, we present a tightly coupled optimization-based GPS-Visual-Inertial odometry system to solve the trajectory drift of the visual-inertial odometry especially over long-term runs. Visual reprojection residuals, IMU residuals, and GPS measurement residuals are jointly minimized within a local bundle adjustment, in which we apply GPS measurements and IMU preintegration used for the IMU residuals to formulate a novel GPS residual. To improve the efficiency and robustness of the system, we propose a fast reference frames initialization method and an online calibration method for GPS-IMU extrinsic and time offset. In addition, we further test the performance and convergence of our online calibration method. Experimental results on EuRoC datasets show that our method consistently outperforms other tightly coupled and loosely coupled approaches. Meanwhile, this system has been validated on KAIST datasets, which proves that our system can work well in the case of visual or GPS failure.

Motivation & Objective

• To address long-term trajectory drift in visual-inertial odometry (VIO) by integrating GPS measurements directly into the optimization framework.
• To improve system robustness and efficiency through online calibration of GPS-IMU extrinsic parameters and time offset.
• To develop a fast reference frame initialization method that accelerates convergence in dynamic environments.
• To validate the system's performance and convergence under visual or GPS signal degradation.
• To demonstrate the effectiveness of a tightly coupled optimization framework combining visual, inertial, and GPS measurements.

Proposed method

• Formulates a novel GPS residual by combining GPS measurements with IMU preintegration to enable tight coupling in the optimization process.
• Performs local bundle adjustment that jointly minimizes visual reprojection errors, IMU preintegration residuals, and the proposed GPS residual.
• Introduces an online calibration method for GPS-IMU extrinsic parameters and time offset using a sliding window optimization framework.
• Employs a fast reference frame initialization technique to accelerate convergence during system startup.
• Uses a tightly coupled optimization framework that fuses visual, inertial, and GPS measurements in a unified cost function.
• Validates the system on EuRoC and KAIST datasets to assess performance under diverse conditions, including sensor failures.

Experimental results

Research questions

• RQ1 Can a tightly coupled optimization framework that fuses GPS, visual, and inertial measurements significantly reduce long-term trajectory drift in VIO?
• RQ2 How effective is the proposed novel GPS residual formulation when combined with IMU preintegration in improving estimation accuracy?
• RQ3 Can online calibration of GPS-IMU extrinsic parameters and time offset enhance system robustness and reduce initialization latency?
• RQ4 How does the system perform under visual or GPS signal degradation, as tested on the KAIST dataset?
• RQ5 What is the convergence behavior and accuracy of the proposed online calibration method in real-world scenarios?

Key findings

• The proposed system consistently outperforms both tightly and loosely coupled approaches on the EuRoC dataset, demonstrating superior trajectory accuracy.
• The system achieves robust performance under visual or GPS signal loss, as validated on the KAIST dataset, confirming resilience to sensor failures.
• The online calibration method converges reliably and improves estimation accuracy by correcting GPS-IMU extrinsic and time offset errors.
• The fast initialization method significantly reduces convergence time compared to conventional initialization techniques.
• The novel GPS residual formulation enhances the system's ability to correct drift over long-term operations.
• The tight coupling of GPS, visual, and inertial measurements leads to more consistent and accurate trajectory estimation than alternative fusion strategies.