xiaochenzhou / fl-datasets-for-har

Four Federated learning datasets for Human Activity Recognition

FL-Datasets-for-HAR

This repo includes four new real-world human activity recognition (HAR) datasets collected under federated learning settings, which first appear at the MobiSys 2021 paper: ClusterFL: A Similarity-Aware Federated Learning System for Human Activity Recognition .

The first dataset is a large-scale dataset collected using an Android App in a crowdsourcing manner. The other three are collected in indoor environments.

Download

The four datasets are publicly available in the current repository. Each dataset is accompanied by a Python file "data_pre.py" for preprocessing and loading each node's data separately in federated learning. The data and processing file are compressed to a ".zip" file for each dataset. Please click the following links for more detail descriptions and downloading each dataset.

HARBox Dataset: Daily Activities Recognition using Smartphones

UWB Dataset: Human Movement Detection using Ultra Wide Band Modules

IMU Dataset: Walking Activity Recognition using Inertial Measurement Unit Modules

Depth Dataset: Gesture Recognition using Depth Camera

Citation

If you find the datasets useful for your research, please cite this paper:

@inproceedings{ouyang2021clusterfl,
  title={ClusterFL: a similarity-aware federated learning system for human activity recognition},
  author={Ouyang, Xiaomin and Xie, Zhiyuan and Zhou, Jiayu and Huang, Jianwei and Xing, Guoliang},
  booktitle={Proceedings of the 19th Annual International Conference on Mobile Systems, Applications, and Services},
  pages={54--66},
  year={2021}
}