Recognition of the Colombian sign language alphabet using Convolutional Neural Networks

Abstract

Sign language provides a system for people with speech or hearing impairments to communicate effectively. However, it is still necessary for the rest of society to appropriate this knowledge. This work consists in designing a computer vision method that recognizes the static signs of the Colombian Sign Language (LSC) alphabet. The methodology consists of a classification algorithm that combines a Convolutional Neural Network (CNN) architecture and image processing techniques. Our approach manages to recognize signs of the alphabet that don’t involve any movement, with 79.2% accuracy. The system is capable of recognizing letters according to the shape, orientation and position of the fingers in each sign, using an imbalanced dataset.