B-HAR-HumanActivityRecognition / B-HAR_Baseline-Human-Activity-Recognition

This package focuses on the definition, standardization, and development of workflow for human activity recognition in depth analysis.

Overview

This repository hosts B-HAR, a baseline framework for in depth study of human activity recognition. B-HAR gives to researchers the possibility to evaluate and compare HAR methodologies with a common well-defined workflow.

Users can interact with B-HAR by using two exposed methods:

• stats
• get_baseline

The stats call return an overview of the dataset under analysis, by giving useful statistics such as data distribution, standard deviation etc.

The get_baseline method takes as input machine learning ad deep learning models (see b_har/utility/models.py) which are going to be applied to classify the input data. It can also a list of patient or activity class to discard from the dataset under analysis.

Other settings such as define the input dataset, workflow, data filtering, etc. can be easily customised in the config.cfg file.

Installation

B-HAR requires python3.6 or higher, you can easily install the package by using pip with the following command:

pip install -i https://test.pypi.org/simple/ B-HAR-baseline-framework

Get started

In order to start using B-HAR you have to follow these two steps:

• Edit the configuration file
• Start the analysis

The code below shows how to use B-HAR.

from b_har.baseline import B_HAR

cfg_file = '/path_to_your_config_file/config.cfg'

b_har = B_HAR(config_file_path=cfg_file)
b_har.stats()
b_har.get_baseline(['K-NN', 'DT', 'LDA'], ['m1_acc'])

You can find a full example and how to use and set the configuration file in the example directory.

Outputs

Once started, B-HAR will create a log directory in which all analysis outputs will be saved. B-HAR will report training stats for both machine learning and deep learning, a sep-by-step recap will be reported in log.rft together with a backup of the configuration file used.

Citing

When using B-HAR please cite the following publication:

Florenc Demrozi, Cristian Turetta and Graziano Pravadelli. "B-HAR: an open-source baseline framework for in depth study of human activity recognition datasets and workflows". https://arxiv.org/abs/2101.10870

Or use:

@misc{demrozi2021bhar,
      title={B-HAR: an open-source baseline framework for in depth study of human activity recognition datasets and workflows}, 
      author={Florenc Demrozi and Cristian Turetta and Graziano Pravadelli},
      year={2021},
      eprint={2101.10870},
      archivePrefix={arXiv},
      primaryClass={eess.SP}
}