PyShoe

Code for "Robust Data-Driven Zero-Velocity Detection for Foot-Mounted Inertial Navigation": an open source foot-mounted, zero-velocity-aided inertial navigation system (INS) that includes implementations of four classical zero-velocity detectors, in addition to our two proposed learning-based detectors. We provide three mixed-motion inertial datasets and evaluation code tools for comparison of the various zero-velocity detection methods.

Dependencies:

numpy
scipy
scikit-learn to run the adaptive zero-velocity detector
pytorch to run the LSTM-based zero-velocity classifier

Scikit-Learn and PyTorch do not need to be installed if you do not intend to use our zero-velocity detectors. You must remove the import of LSTM and SVM from ins_tools/EKF.py to do so.

Datasets

Download the full inertial dataset (~2.9 GB) by running the following bash script from the source directory (to automatically download and extract the dataset):

bash download_data.sh

We currently provide three separate inertial datasets in the full package:

VICON Dataset: Stored in data/vicon. Collected within a ~3x3m motion capture area. There are 56 trials in total, and all of the raw data has been processed into a .mat file in data/vicon/processed. This dataset has complete position ground truth.
Hallway Dataset: Stored in data/hallway. Consists of walking and running motions along three hallways (39 trials in total). There are intermediate ground truth positions along the path: we provide the time steps (indices of the IMU sequence) that correspond with the ground truth positions.
Stair Dataset: Stored in data/stairs. Consists of stairway trials (six training trials and eight testing trials). There is intermediate vertical position ground truth for each flight: we provide the time steps (indices of the IMU sequence) that correspond with these locations.

Zero-Velocity-Aided INS

We have open-sourced our Python-based zero-velocity-aided INS that uses an error-state extended Kalman filter to fuse zero-velocity pseudo-measurements that are produced by a zero-velocity detector. We refer to our papers and the citations therein for more technical details about the INS. We would like to acknowledge openshoe; we based some components of our zero-velocity-aided INS on their open-source MATLAB implementation.

We have included a script, ins_demos.py, with five separate demos - this sample code shows how to run our INS with data from the three datasets that are provided. Two of the demos show how our proposed learning-based zero-velocity detectors can be used.

Reproduction of Paper Results

We have included two scripts that reproduce the main results in our paper:

process_error_hallway.py

This script reproduces Table 3 of the paper by iteratively evaluating the error of specified zero-velocity detectors (using any specified list of zero-velocity thresholds) for all trials in the hallway dataset. Two CSV files are automatically generated in the results folder: hallway_results_raw.csv, which shows the error for all individual trials, and hallway_results.csv, which is a processed version that reproduces the paper results. The processed results are saved to results/hallway_results.csv.

plot_hallway_data.py

This script generates plots for all of the hallway trials within results/figs/hallway/.

process_error_stairs.py

This script reproduces Table 4 of the paper by iteratively evaluating the error of the specified zero-velocity detectors (each with a list of threshold values) for all trials in the stair dataset Two CSV files are automatically generated in the results folder: stair_results_raw.csv, which shows the error for all individual trials, and stair_results.csv, which is a processed version that reproduces the paper results. The processed results are saved to results/stair_results.csv.

plot_stair_data.py

This script generates plots for all of the hallway trials within results/figs/hallway/.

Additional Scripts

run_threshold_optimization: Generates the zero-velocity labels that we used to train our LSTM-based zero-velocity classifier. The zero-velocity thresholds for the five zero-velocity detectors (SHOE, ARED, AMVD, MBGTD, VICON) are optimized for each VICON dataset trial, and the most accurate detector's output is used as the ground truth label for each trial.
process_error_vicon.py: Iteratively evaluates the error of the five classical zero-velocity detectors when their optimized zero-velocity outputs are used. The results are saved in results/vicon_results_raw.csv.
train_motion_classifier.py: Trains a three-class motion classifier (walk vs. run vs. stairs) using a subset of the training data, and will reproduce the accuracy of the classifier for the validation set.

Citation

If you use this code in your research, please cite:

@article{2019_Wagstaff_Robust,
  title = {Robust Data-Driven Zero-Velocity Detection for Foot-Mounted Inertial Navigation},
  author = {Brandon Wagstaff and Valentin Peretroukhin and Jonathan Kelly},
  journal = {IEEE Sensors Journal},
  pages = {957--967},
  number = {2},
  url = {https://arxiv.org/abs/1910.00529},
  volume = {20},
  year = {2019}
}

@inproceedings{2018_Wagstaff_LSTM-Based,
  address = {Nantes, France},
  author = {Brandon Wagstaff and Jonathan Kelly},
  booktitle = {Proceedings of the International Conference on Indoor Positioning and Indoor Navigation {(IPIN'18)}},
  date = {2018-09-24/2018-09-27},
  month = {Sep. 24--27},
  title = {LSTM-Based Zero-Velocity Detection for Robust Inertial Navigation},
  url = {http://arxiv.org/abs/1807.05275},
  year = {2018}
}

@inproceedings{2017_Wagstaff_Improving,
  address = {Sapporo, Japan},
  author = {Brandon Wagstaff and Valentin Peretroukhin and Jonathan Kelly},
  booktitle = {Proceedings of the International Conference on Indoor Positioning and Indoor Navigation {(IPIN'17)}},
  date = {2017-09-18/2017-09-21},
  doi = {10.1109/IPIN.2017.8115947},
  month = {Sep. 18--21},
  title = {Improving Foot-Mounted Inertial Navigation Through Real-Time Motion Classification},
  url = {http://arxiv.org/abs/1707.01152},
  year = {2017}
}