Applying Singular Value Decomposition on Accelerometer Data for 1D Convolutional Neural Network Based Fall Detection

This is the code for the 2019 Electronics Letters paper Applying Singular Value Decomposition on Accelerometer Data for 1D Convolutional Neural Network Based Fall Detection by Heeryon Cho and Sang Min Yoon.

Requirements

This code runs with:
* Ubuntu 16.04
* NVIDIA GeForce GTX 960
* TensorFlow version 1.5.0
* Python 2.7.12

Downloading Fall Recognition Datasets

Please download the following three Human Activity Recognition benchmark datasets (which include fall activities) from the respective sites and place them inside the 'raw_dataset' folder.
* SisFall (http://sistemic.udea.edu.co/en/research/projects/english-falls/)
* UMAFall (https://figshare.com/articles/UMA_ADL_FALL_Dataset_zip/4214283/6) [version 6]
* UniMiB (http://www.sal.disco.unimib.it/technologies/unimib-shar/)

Remarks

1. Place the unzipped benchmark dataset into the 'raw_dataset' folder. Refer to the directory structure given in the 'raw_dataset' folder's readme.txt. (Note: For UniMiB data, you first need to convert the .mat files to .csv files using the 'convert_mat2csv.py' code located inside the UniMiB folder.)
2. Generate processed (raw or dimension reduced) data by executing codes marked 'gen_XXX.py'. A sample data, generated using the code 'umafall/gen_umafall_dataset_kpca.py', is given inside 'umafall/data/' folder (Two pickle files: X_umafall_kpca.p & y_umafall_kpca.p).
3. Build and evaluate 1D CNN models. A sample model, generated using the code 'umafall/umafall_kpca_conv1d_10.py', is given inside 'umafall/model/umafall_kpca_conv1d_10/' folder.

Citation

If you find this useful, please cite our work as follows:

@article{ChoYoon_2019ElectronicsLetters,
  author    = {Heeryon Cho and Sang Min Yoon},
  title     = {Applying Singular Value Decomposition on Accelerometer Data for 
               1D Convolutional Neural Network Based Fall Detection},
  journal   = {Electronics Letters},
  volume    = {55},
  number    = {6},
  pages     = {320--322},
  doi       = {10.1049/el.2018.6117},
  year      = {2019},
}