This repository is an artifact for the paper "CNNs for Heart Rate Estimation and Human Activity Recognition in Wrist Worn Sensing Applications" submitted to the WristSense workshop as part of PerCom 2020.
The code has been supplied as Jupyter Notebooks and designed to run in Google Colaboratory. The dataset used is open source and freely available. The data was collected by D. Jarchi and A. Casson (2017) and downloaded from PhysioNet.
You should execute the repo in the following order:
1. Create a directory in Google Drive named 'WristSense'
2. Clone this GitHub repository into 'WristSense'
3. Run the notebook Download_Data.ipynb
4. Run the notebook CNN_Recurrent.ipynb
5. Run the notebook HAR_Data.ipynb
6. Run the notebook Transfer_Learning_HAR.ipynb
Download_Data.ipynb
This notebook will download the data necessary to complete the experiments below. It will also create some directories to store your data and results
CNN_Recurrent.ipynb
When you call the RCNN function you can specify Conv-Pooling params which will affect the outcome of your heart rate error.
Your choice of conv-pooling filter (cv_k) and stride (cv_k) sizes will be dependent on seq_len that changes with you downsampling factors dwns_factor. You can set these in the call_RCNN() function.
The results will be written to a json file in format:
[batch_size, exercise, downsampled frequency, heart rate error]
HAR_Data.ipynb
This notebook provides functions to segment, plot and save the individual images used in the Tranfer Learning section of this experiment.
Transfer_Learning_HAR.ipynb
This notebook computes Transfer Learning on a Inception-v3 model pretrained on ImageNet.
If you find this repo helpful in any way please cite our forthcoming paper:
@inproceedings{Brophy2020,
author = {Brophy, Eoin and Muehlhausen, Willie and Smeaton, Alan F. and Ward, Tomas},
title = {CNNs for Heart Rate Estimation and Human Activity Recognition in Wrist Worn Sensing Applications},
booktitle = {2020 {IEEE} International Conference on Pervasive Computing and Communications,
PerCom, Austin, Texas, March 23-27, 2020},
publisher = {{IEEE Computer Society}},
year = {2020}
}
No longer necessary -> You should include the following command at the top of your .bib file to cite a forthcoming paper: @preamble{ " \newcommand{\noop}[1]{} " }