Development of a Wearable Electrical Impedance Tomographic Sensor for Gesture Recognition With Machine Learning

Area of Science:

• Biomedical Engineering
• Electrical Engineering
• Machine Learning

Background:

• Wearable electrical impedance tomography (wEIT) offers a promising approach for non-invasive human-computer interaction.
• Optimizing sensor design and algorithms is crucial for enhancing the accuracy of gesture recognition systems.

Purpose of the Study:

• To develop and optimize a wearable electrical impedance tomographic (wEIT) sensor for gesture recognition.
• To evaluate the impact of electrode materials and shapes on recognition rates.
• To identify the most effective machine learning algorithms for classifying gestures using wEIT data.

Main Methods:

• A wearable sensor with 8 electrodes was designed and tested.
• Various electrode materials and shapes were compared to determine optimal configurations.
• Multiple machine learning algorithms were applied to voltage data from three distinct gestures.
• An electrical model of electrode-skin contact impedance was established to understand the recognition mechanism.

Main Results:

• Rectangular copper electrodes yielded the highest gesture recognition rates.
• Electrode-skin contact impedance was found to improve recognition accuracy.
• The Medium Gaussian Support Vector Machine (SVM) algorithm demonstrated superior performance, achieving an average recognition rate of 95%.

Conclusions:

• The developed wEIT sensor, utilizing rectangular copper electrodes and the Medium Gaussian SVM algorithm, effectively recognizes gestures with high accuracy.
• Understanding and modeling electrode-skin contact impedance is vital for improving wEIT-based gesture recognition.
• This research contributes to the advancement of wearable sensing technologies for human-computer interaction.