On the Selection of Number of Sensors for a Wearable Sign Language Recognition System
Rinki Gupta · 2019
Sign language is the primary language used by the deaf community. The postures and motion of hands of the signer may be captured using wearable sensors to design an electronic interface between a sign language user and a non-signer. In a wearable sign language recognition system, the signals recorded using the sensors are analyzed for identifying which sign is performed. For an accurate and reliable sign language recognition system, the primary requirement is to be able to record such signals that will be able to identify two different signs as distinct. With advancements in sensor manufacturing, surface electromyograms (sEMG) have increasingly been reported in research on human motion analysis. In this paper, an sEMG based system is explored for recognition of basic hand postures and motions used in the Indian Sign Language. A backward search algorithm is proposed to determine the most suitable number and placement of sensors based on their effect on misclassification error. A percentage misclassification error of 8.9% is obtained with 5 sEMG sensors, which degrades as the number of sensors are reduced and reaches up to 30% when only one sensor is used. There is a tradeoff between the classification accuracy and the cost of the system.