Four Sensors Bracelet for American Sign Language Recognition based on Wrist Force Myography

Abstract

Recently, Force myography (FMG) has emerged as an alternate method for gesture recognition applications. It is usually used as raw signal, in combination with data acquired from other sensors or with a large number of sensors for an efficient recognition performances. Only four FMG commercial sensors constitute the proposed gesture recognition system. They are connected as an FSR sensor’s bracelet applied for american sign language(ASL) recognition. In this paper, a comparative study between raw FMG and six commonly extracted features when implemented on the Extreme Learning Machine (ELM) to assess the accuracy of nine ALS alphabet recognition system. 16 trials of the nine gestures were collected from a healthy male wearing the FMG bracelet. In addition, 5 folds cross validation was implemented during the ELM training. As results it was noted that the accuracy based on six commonly features was equal to 89.65%, which over perform not only the raw FMG based gesture recognition that reached only a testing total accuracy of 68.96% by our 4 sensors but also some 8 sensor’s raw data systems in literature.