Recognition of the Hungarian fingerspelling alphabet using Recurrent Neural Network

Abstract

The aim of this paper is to introduce a Recurrent Convolutional Neural Network based on depth data to recognize the signs of the Hungarian fingerspelling alphabet. The training dataset contains depth data of 27 static and 15 dynamic signs. A 88.6% classification accuracy was measured for during the test with the recommended model in this paper, which is a special type of recurrent network containing LSTM and convolutional layers.