Investigation of the effects of different arm positions and angles in sEMG-based hand gesture recognition on classification success

The effective operation of surface electromyography (sEMG) signal-based controlled active prostheses and human-machine interaction systems in daily life is crucial to work with high accuracy in different angles and positions of the arm. In this study, sEMG recordings from three different positions and angles of the arm were combined with accelerometer and gyroscope data to classify four different hand movements. The classification data (8-channel sEMG, accelerometer, and gyroscope) were collected from the right forearm of 13 participants. To create the dataset, six features were extracted from sEMG signals and three from accelerometer and gyroscope data. As a result, the methodological investigation was carried out on how different arm positions and angles affect the classification of hand movements. Evaluations were also made regarding whether the adverse effects arising from different arm positions and angles of the movement could be mitigated using accelerometer and gyroscope data, and their effects on classifier performance were discussed. As classifiers, Artificial Neural Networks (ANN) and Support Vector Machines (SVM) were used. SVM classifiers achieved an average success rate of 83% in the five different categories of analysis, while ANN classifiers achieved an average success rate of 82%. It was found that accelerometer and gyroscope data in various positions contributed very little to the performance of movement classification. As a result of the evaluation, it was discovered that collecting training data in all positions and angles of the forearm improved classification results for a sEMG-based systems.