Pedestrian Inertial Navigation with Multi-Head CNN

Click here to access the paper.

Fig 1. Pedestrian gait cycle

Fig 2. Proposed Multi-Head CNN (MHCNN) prediction model

Fig 3. Pedestrian trajectory estimations

The error-state KF pedestrian INS (MATLAB) implementation can be found at OpenShoe webpage.

If you like to cite this work, please use the BibTeX info

@INPROCEEDINGS
{
    PIN-MHCNN,
    author={Cetin, Gokhan and Kucuk, Mehmet Ali and Koroglu, Muhammed Taha},
    booktitle={2023 IEEE International Workshop on Metrology for Industry 4.0 & IoT (MetroInd4.0&IoT)}, 
    title={Pedestrian Inertial Navigation with Multi-Head CNN}, 
    year={2023},
    volume={},
    number={},
    pages={275-280},
    doi={10.1109/MetroInd4.0IoT57462.2023.10180130}
}

or the following plain text.

G. Cetin, M. A. Kucuk and M. T. Koroglu, "Pedestrian Inertial Navigation with Multi-Head CNN," 2023 IEEE International Workshop on Metrology for Industry 4.0 & IoT (MetroInd4.0&IoT), Brescia, Italy, 2023, pp. 275-280.

The trained network (the file with h5 extension) is not in the repo: Github rejected uploading the model file due to large size and while using rm command in git to remove the model file from the added files, the model is accidentally deleted. Yet, one can reproduce it by training the model from scratch (just run code/mhcnn-training/imu-localization.ipynb - training takes approximately 11 hours with the PC mentioned in the paper).