The Invisible Map: Visual-Inertial SLAM with Fiducial Markers for Smartphone-based Indoor Navigation

Paul Ruvolo,Ayush Chakraborty,Rucha Dave,Richard Li,Duncan Mazza,Xierui Shen,Raiyan Siddique,Krishna Suresh

2023-10-17

Abstract:We present a system for creating building-scale, easily navigable 3D maps using mainstream smartphones. In our approach, we formulate the 3D-mapping problem as an instance of Graph SLAM and infer the position of both building landmarks (fiducial markers) and navigable paths through the environment (phone poses). Our results demonstrate the system's ability to create accurate 3D maps. Further, we highlight the importance of careful selection of mapping hyperparameters and provide a novel technique for tuning these hyperparameters to adapt our algorithm to new environments.

Robotics,Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: in an indoor environment, how to use mainstream smartphones to create large - scale, easy - to - navigate 3D maps while achieving precise positioning and navigation. Specifically, the author focuses on how to combine Visual - Inertial Odometry (VIO) and fiducial markers (such as AprilTags) to solve the Simultaneous Localization and Mapping (SLAM) problem.

### Problem Background

1. **Importance of Indoor 3D Mapping and Navigation**
- For autonomous mobile robots and pedestrians, it is crucial to accurately position and plan the optimal path in complex indoor environments.
- Especially for the visually impaired or people navigating in large and unfamiliar environments, indoor navigation is becoming increasingly important.

2. **Limitations of Existing SLAM Methods**
- Many SLAM studies focus on wheeled mobile robots, using data fusion of wheel encoders and ranging sensors.
- Although the VIO algorithms on modern smartphones are relatively accurate, they can only provide relative motion estimates, and errors will accumulate over time, resulting in large - scale positioning failures.

### Solution

To solve the above problems, the author proposes a fiducial - marker - based SLAM system, with the following main contributions:

1. **Optimized Fiducial - Marker - Based SLAM System**
- Using the Graph SLAM framework, combine VIO data with fiducial - marker observations to generate accurate 3D maps.
- Fiducial markers (such as AprilTags) can be accurately detected by smartphone cameras under different lighting conditions, providing an absolute position reference to make up for the deficiencies of VIO.

2. **Importance of Hyper - Parameter Tuning**
- Explore the impact of hyper - parameter selection on navigation effects, and provide methods for hyper - parameter tuning with and without ground - truth datasets.
- Introduce the "shift metric" as a new index for evaluating map quality, which is especially suitable for hyper - parameter tuning in new environments.

3. **Publicly Available Smartphone Application**
- Deploy this SLAM system into a publicly available smartphone application for map creation and navigation.

### Method Overview

- **Map Generation**: The user captures the positions of fiducial markers through the mobile phone camera and inputs the information of points of interest in the environment to generate time - series data containing device poses and landmark poses.
- **User Localization**: Use a similar map - generation optimization formula, fix the landmark positions, and update the user's current position estimate.
- **Shortest - Path Planning**: Represent the generated line segments as a weighted graph, detect self - intersections and insert additional nodes as connection points to generate the shortest path from the current position to the target position.

Through these methods, the author aims to provide a robust and easy - to - use indoor navigation system, especially suitable for the visually impaired or other people in need of assisted navigation.