An Open-Source Data Acquisition and Manual Segmentation System for Hand Gesture Recognition based on EMG

Abstract

Due to lack of standardization in the data acquisition process, Hand Gesture Recognition literature has produced a high number of different but incompatible datasets. This paper presents a system for data acquisition of EMG signals and its manual segmentation. The system can be connected with the two most affordable wearable EMG armbands: Myo Armband and gForce Pro. The system allows to record a given number of samples per gesture during a given number of seconds. Twelve gestures were selected for being natural and the most reported in the literature. The system includes several features that enhance the quality of the dataset such as: strategies to maintain the volunteer attention, and the capability to resume recording in case of interruption. The system was evaluated using the Computer System Usability Questionnaire (CSUQ) over 10 data collectors. This questionnaire allowed to obtain System quality (85.5 %), Information quality (84.5 %) and Interface quality (89.5%) perceptions with an overall usability of 85.9%. These results show that the system is greatly designed, intuitive and of ease of use. The software is publicly available and was developed in Matlab.