Towards an Electromyographic Armband with dry electrodes for Hand Gesture Recognition

Gesture recognition is a computer discipline based on mathematical algorithms for the interpretation of human gestures. In recent years, recognizing and identifying hand movements have gained a lot of interest due to the increasing of Human-Computer Interface (HCI) applications like the control of mobile apps, robotics, video games, and, in the clinical field, prosthesis and robotic limb control. Several techniques are currently available for data collection: this thesis focuses on the acquisition of the surface ElectroMyoGraphic (sEMG) signal, by applying noninvasive electrodes on the skin above the muscle of interest. This thesis work studies the implementation of dry electrodes for sEMG acquisition, evaluating signal quality by means of the Signal-to-Noise Ratio (SNRdB). An acceptance value of 10 dB has been fixed, in order to consider the signal suitable for an event-based extraction technique. In particular, the final goal is to develop an embedded low-power system for gesture recognition based on the Average Threshold Crossing (ATC) technique, which is performed by establishing a threshold above the sEMG signal at rest condition and averaging the number of over-crossings inside a fixed time window. With the aim to replace wet with dry electrodes, a first analysis involves three types of dry ones. Dry electrodes with the highest SNRdB are selected for the subsequent EMG acquisitions. The sEMG signals are detected using a custom PCB developed by the research group and they are digitalized by means of a Data Acquisition System (DAQ) for further processing. To increase the SNRdB a Driven Right Leg (DRL) circuit available onboard was tested in order to reduce the common-mode noise on the skin and match the body’s reference potential to the circuit’s reference voltage. To allow an embedded application, the research group has developed an improved PCB characterized by a MicroController Unit (MCU) onboard. The MCU permits to transfer data via Bluetooth, so there is an incrementation of SNRdB because the board’s connections reduce the noise previously introduced through the DAQ system. This configuration was tested by means of procedures similar to the ones of the previous version, adding a comparison with wet electrodes. Finally, a 3D model of the armband has been designed, composed of seven acquisition channels secured together by a plastic strip. One package was effectively prototyped to perform the tests. The SNRdB calculated averaging all the contraction was equal to ~22 dB for the acquisition with wet electrodes and to ~17 dB for dry ones. The obtained SNR revealed a good quality of the acquired sEMG signal, thus confirming the feasibility of an armband for ATC-based hand gesture recognition.