HGRLab

HGRLab: Open-Source Toolkit for Hand Gesture Recognition

We suggest using a virtual environment to install the necessary dependencies and run the code. To install the required package dependencies using pip, please run the following command:

pip install -r requirements.txt

Please note that currently, this project does not offer an installation package. Therefore, all commands must be executed from the project root directory hgrlab.

Comparison of sEMG-based Hand Gesture Classifiers

The source code of the experiment delello_lnlm2024 allows the reproducibility of the results presented in the paper entitled "Comparison of sEMG-based Hand Gesture Classifiers" authored by Guilherme C. De Lello, Gabriel S. Chaves, Juliano F. Caldeira, and Markus V.S. Lima.

The experiment aims to analyze the effect of different classifiers on hand gesture classification. The study considers five supervised learning classifiers, namely support vector machine, logistic regression, linear discriminant analysis, k-nearest neighbors, and decision tree.

For this experiment, the model hgr_dtw was used, whose architecture is based on the HGR system described in the paper Real-Time Hand Gesture Recognition Based on Artificial Feed-Forward Neural Networks and EMG by Benalcázar et al.

The k-fold cross-validation method was employed to estimate the threshold that optimizes individual accuracy for all subjects, instead of using a fixed muscle detection threshold. This process is explained in the paper Hand Gesture Classification using sEMG Data: Combining Gesture Detection and Cross-Validation by Chaves et al.

To run this experiment, please follow these steps:

1. Download the project source code
2. Open your terminal application
3. Make sure the required packages are installed
4. Set the current directory to the project root directory hgrlab
5. Run the command python -m hgrlab.experiments.delello_lnlm2024

This program performs the following tasks:

• Download the HGR datasets used in the experiment, including sEMG data from 10 subjects
• Experiment 1: Determine the best segmentation thresholds for each of the 10 subjects and 5 classifiers using 4-fold cross-validation (Table 1: Optimum individual segmentation thresholds using 4-fold cross-validation)
• Experiment 2: Compare the accuracy of the HGR system for each of the 5 classifiers (Table 2: Mean accuracy and standard deviation of the HGR systems using different classifiers and Table 3: Mean accuracy and standard deviation by subject for different classifiers)

If you use HGRLab in a scientific publication, we would appreciate citations to the following paper:

Comparison of sEMG-based Hand Gesture Classifiers, De Lello et al., Learning and Nonlinear Models, vol. 22, no. 2, pp. 48–61, October 2024.

Bibtex entry:

@article{de_lello_comparison_2024,
    title = {Comparison of {SEMG}-{Based} {Hand} {Gesture} {Classifiers}},
    author = {De Lello, G. C. and Chaves, G. S.
             and Caldeira, J. F. and Lima, M. V. S.},
    journal = {Learning and Nonlinear Models},
    volume = {22},
    number = {2},
    pages = {48--61},
    year = {2024},
    month = oct,
    doi = {10.21528/lnlm-vol22-no2-art4},
}