A Comprehensive Introduction of Visual-Inertial Navigation

Yangyang Ning

2023-06-28

Abstract:In this article, a tutorial introduction to visual-inertial navigation(VIN) is presented. Visual and inertial perception are two complementary sensing modalities. Cameras and inertial measurement units (IMU) are the corresponding sensors for these two modalities. The low cost and light weight of camera-IMU sensor combinations make them ubiquitous in robotic navigation. Visual-inertial Navigation is a state estimation problem, that estimates the ego-motion and local environment of the sensor platform. This paper presents visual-inertial navigation in the classical state estimation framework, first illustrating the estimation problem in terms of state variables and system models, including related quantities representations (Parameterizations), IMU dynamic and camera measurement models, and corresponding general probabilistic graphical models (Factor Graph). Secondly, we investigate the existing model-based estimation methodologies, these involve filter-based and optimization-based frameworks and related on-manifold operations. We also discuss the calibration of some relevant parameters, also initialization of state of interest in optimization-based frameworks. Then the evaluation and improvement of VIN in terms of accuracy, efficiency, and robustness are discussed. Finally, we briefly mention the recent development of learning-based methods that may become alternatives to traditional model-based methods.

Robotics

What problem does this paper attempt to address?

The paper primarily explores the issue of Visual-Inertial Navigation (VIN) and provides a comprehensive introduction to the related theories and technologies. Specifically, the paper aims to address the following key issues:

1. **Fusion of Visual and Inertial Sensing**: How to effectively combine data from cameras (visual sensors) and Inertial Measurement Units (IMUs) for navigation, in order to estimate the ego-motion of the sensor platform and its local environment.

2. **State Estimation Framework**: The paper introduces the VIN problem within the classical state estimation framework, including state variables, system models, and related parameter representations (such as parameterization methods), IMU dynamics models, camera measurement models, etc., and discusses the corresponding probabilistic graphical models (factor graphs).

3. **Existing Estimation Methods**: The study examines existing model-based estimation methods, including methods under filtering and optimization frameworks and related operations. It also discusses parameter calibration and initialization issues within the optimization framework.

4. **Performance Evaluation and Improvement**: The paper further discusses strategies for evaluating and improving VIN in terms of accuracy, efficiency, and robustness.

5. **Development of Learning-Based Methods**: Finally, the paper mentions that learning-based methods in recent years may become an alternative to traditional model-based methods.

In summary, the main goal of this paper is to provide a comprehensive tutorial introduction to Visual-Inertial Navigation, covering various aspects from basic theory to practical applications, aiming to provide researchers and engineers with a resource for in-depth understanding of the VIN field.