Sign language numeral gestures recognition using convolutional neural network

Abstract

This paper presents usage of convolutional neural network for classification of sign language numeral gestures. For requirements of this research, we created a new dataset of these gestures. The dataset was recorded via Kinect v2 device and it consists of recordings of 18 different people. Only depth data-stream was used in our research. For a classification task, there was utilized classic VGG16 architecture and its results were compared with chosen baseline method and other tested architectures. Our experiment on classification showed the great potential of neural networks for this task. We reached recognition accuracy 86.45%, which is by more than 34% better result than chosen baseline method.