Rock beats Scissor: SVM based gesture recognition with data gloves

Abstract

Hand gestures play an important role in human communication, particularly when auditory communication is limited. Akin to speech recognition, hand gesture recognition can therefore be a useful tool to facilitate communication and for more immersive computer interaction. In this paper, we examine the mobile recognition of hand gestures using data recorded with sensor gloves. We design a system based on Support Vector Machines (SVM), capable of recognizing 5 different hand gestures. In an experiment with 11 participants, we determine applicable hyperparameters based on performance on the training set which translates into 100% classification accuracy on the test set. In an additional practical experiment with 9 participants, our system achieves up to 98% in a personalized and up to 87.5% in a generalized model setting.