InCrowd-VI: A Realistic Visual–Inertial Dataset for Evaluating SLAM in Indoor Pedestrian-Rich Spaces for Human Navigation

Paper | Dataset | Support

Example scenes from the InCrowd-VI dataset demonstrating various challenges: (a) high pedestrian density, (b) varying lighting conditions, (c) texture-poor surfaces, (d) reflective surfaces, (e) narrow aisles, and (f) stairs.

About

A realistic visual-inertial dataset with 58 sequences spanning 5km of trajectories and 1.5 hours of recordings, designed for evaluating SLAM systems in indoor pedestrian-rich environments. The dataset is particularly aimed at advancing navigation technologies for visually impaired individuals.

For more details, please refer to the accompanying published paper.

File structure and documentation

Details on the file structure for each sequence are available on the Sequence file structure page. Additionally, you can find related documentation in the repository:

• Calibration data: Calibration parameters for cameras and IMUs used in the dataset.
• InCrowd-VI sequences: Overview of sequences with density categories and main challenges.

Due to the large size of image files, the dataset provides .vrs files for each sequence. You can extract the necessary images locally using the tools available in this repository.

---

Getting started

Requirements

The dataset generation tools have been tested on the following system configuration:

• Operating system: Ubuntu 22.04 LTS
• Python version: 3.10
• System dependencies:

sudo apt install python3.10-venv

To extract images from .vrs files, the projectaria_tools Python package is required. For detailed information about the package, refer to the official Meta documentation:

• Getting Started with Data Utilities
• Installation Instructions for Python

Step 1. Setting up a virtual environment

To manage dependencies:

python3 -m venv $HOME/projectaria_tools_python_env
source $HOME/projectaria_tools_python_env/bin/activate

⚠️ IMPORTANT: Ensure the virtual environment is activated before running any dataset generation scripts.

Step 2. Install projectaria_tools

Upgrade pip and install the package:

python3 -m pip install --upgrade pip
python3 -m pip install projectaria-tools'[all]'

Dataset download

The dataset sequences are publicly available at: InCrowd-VI Dataset

Extract data

The dataset was collected with Meta Aria Project glasses worn by a walking person. Aria data is recorded using VRS, an open-source file format. To extract data from .vrs files, use the provided scripts located in the tools/dataset_generation directory. These scripts process the .vrs file to undistort images and extract stereo and RGB images.

Required scripts

Ensure you download the following scripts from the tools/dataset_generation directory:

• bb_generate_dataset_in_loop.sh: Processes multiple .vrs files sequentially using the bb_generate_sequence_ns.sh script.
• bb_generate_sequence_ns.sh: Extracts and undistorts images from a .vrs file, organizes them by sensor type and timestamp unit, and generates a timestamps.txt file.
• bb_image_undistortion.py: Processes .vrs files to extract and undistort RGB and stereo images, saving both original and undistorted images with timestamps as filenames.

These scripts must be placed in the same directory before proceeding.

Steps to extract data

1. Download the scripts

Download the required scripts from the repository's tools/dataset_generation directory. Place them in the desired directory.

2. Select .vrs files for extraction

Download the .vrs files from the InCrowd-VI Dataset. Then, open the bb_generate_dataset_in_loop.sh script and modify the files=(...) section. Uncomment the block corresponding to the vrs files you want to process.

3. Run the data generation script

Execute the script:

bash bb_generate_dataset_in_loop.sh

⚠️ IMPORTANT: Ensure the virtual environment is activated before running this script.

⚠️ IMPORTANT: Ensure that only ONE files=(...) block is uncommented at a time to avoid unintended behavior.

Citation

If you use this dataset or refer to the paper, please cite it as follows:

@article{bamdad2024incrowd,
  title={InCrowd-VI: A Realistic Visual--Inertial Dataset for Evaluating Simultaneous Localization and Mapping in Indoor Pedestrian-Rich Spaces for Human Navigation},
  author={Bamdad, Marziyeh and Hutter, Hans-Peter and Darvishy, Alireza},
  journal={Sensors},
  volume={24},
  number={24},
  pages={8164},
  year={2024},
  publisher={MDPI}
}

Acknowledgments

Special thanks to the Robotics and Perception Group at the University of Zurich for providing the Meta Aria glasses used in this study.

Support

If you have any questions or encounter issues, please create an issue in this repository. We will address it as soon as possible.