Data Acquisition Software for Sign Language Recognition

Abstract

The research of sign language recognition is a topic of interest in the recent years in the fields of computing vision, artificial intelligence and human-computer interaction. The sign language recognition has at least three steps such as data acquisition, data classification and results. This paper introduces a software capable of facilitating the data acquisition step with the Intel Re-alsense camera. This software digitize 22 points from the hand of the user in coordinates (x,y,z) and save this data in a comma separated values (CSV) file. Finally, in this paper we tested the software with 19 users recording the alphabet of the Mexican sign language, with the dataset from the software, we trained an artificial neural network to obtain 80.1192% of precision. While this percentage can be increased with a variety of techniques, the test shows that users can create the dataset using this software and that those values can be used in a classification algorithm with acceptable results.