Embedded System for Hand Gesture Recognition Using EMG Signals: Effect of Size in the Analysis Windows

The electromyography signals (EMG) analysis in the field of robotics has had a great impact due to its application in prosthesis and system control using electrical signals resulting from the action of muscles associated with different parts of the human body. In this article, an embedded system of hand gesture recognition based on EMG signals measured in the forearm is implemented. The EMG signals are acquired through the Myo Armband sensor and processed with a 32-bit STM microcontroller. The signals are classified with the Naïve Bayes algorithm (NB) and by recognizing an established pattern (open or closed), the system sends control signals to a robotic hand to replicate the movement. As main contribution, a comparative analysis of the performance of the embedded system is presented based on: number of samples (analysis windows), acquisition and processing times, as well as the macro and micro-evaluation of multiclass metrics (Accuracy, Precision, Recall and F-Score) for the recognition of movements.