Modelo de reconocimiento para la lengua de señas: aproximación comparativa entre métodos de reconocimiento de patrones por inteligencia artificial
Thesis
The sign language is the communication tool that is most used within the hearingimpaired people community, as it allows their users to communicate through gestures and movements. Even though, in Colombia and some other countries in the world as India and the U.S, the challenge with education, standardization and teaching of this language is evident, an example of this are the multiple variations on these languages between the different cultures and geographic zones. For this reason, the methodologies that allows the automatization of the teaching and communication process within the users of this language (even if they are hearing impaired or not), are relevant to accomplish the inclusion within a social and educational context for the deaf people and those with any type of hear impairing. In this order this investigation aims to study alternatives as algorithms based on Neural Networks and Machine Learning, to generate a model that can recognize and classify different hand gestures part of the alphabet from the American Sign Language (ASL). All the mentioned before will be done training and validating three initial models based on Convolutional Neuronal Networks (CNN) which will be explored systematically with adjustments on structure and hyper-parameters to identify the model structure that adapts the better to the appropriate classification of each of the 27 types of images part of the signs on the ASL alphabet.
