Finger language recognition based on ensemble artificial neural network learning using armband EMG sensors

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Human-Computer Interaction

Background:

• Deaf individuals face communication barriers, leading to social and financial disadvantages.
• Specialized communication methods like sign and finger languages can limit interaction.
• Developing assistive technologies is crucial for inclusivity.

Purpose of the Study:

• To develop a novel finger language recognition algorithm.
• To utilize an ensemble artificial neural network (E-ANN) for enhanced accuracy.
• To employ an armband system with electromyography (EMG) sensors for data acquisition.

Main Methods:

• Signal acquisition, filtering, segmentation, and feature extraction were performed.
• An E-ANN classifier was trained and evaluated using Korean finger language data.
• Performance was assessed using 5-fold cross-validation and compared against a standard artificial neural network (ANN).

Main Results:

• Increasing the number of E-ANN classifiers and training data size improved average accuracy.
• Higher classifier counts (up to 8) and data size (300) reduced accuracy variability.
• The optimal E-ANN configuration demonstrated significantly higher accuracy than a general ANN.

Conclusions:

• The developed E-ANN finger language recognition system offers superior performance.
• Optimized E-ANN models enhance accuracy and reliability in recognizing finger gestures.
• This technology holds potential for reducing communication barriers for the deaf community.