Improvements on EMG-based handwriting recognition with DTW algorithm.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Electromyography (EMG) signals are increasingly used for human-computer interaction.
• Dynamic Time Warping (DTW) is a recognized technique for feature extraction in EMG-based handwriting recognition.
• Existing methods may have limitations in classification accuracy and interclass variance.

Purpose of the Study:

• To improve the classification process and enhance the recognition accuracy of EMG-based handwriting recognition.
• To introduce modifications to the Dynamic Time Warping (DTW) algorithm and distance metrics.
• To reduce interclass variance and generate more salient features for better classification.

Main Methods:

• A two-phase template making approach was developed to create templates with enhanced features.
• Modified Mahalanobis Distance (mMD) was employed to replace Euclidean Distance (ED) for improved classification.
• Experiments involved four subjects writing lowercase letters, with four-channel EMG signals recorded from the forearm.

Main Results:

• The proposed modifications led to a significant increase in average recognition accuracy.
• Offline analysis demonstrated an improvement of 9.20% in average recognition accuracy.
• The new methods effectively minimized interclass variance, enhancing classification performance.

Conclusions:

• The enhanced DTW feature extraction and mMD classification methods offer a substantial improvement for EMG-based handwriting recognition.
• The two-phase template making approach contributes to more robust feature representation.
• These advancements pave the way for more accurate and reliable EMG-driven interfaces.