Diseño e implementación de un sistema de traducción de lenguaje de señas dactilológico a lenguaje natural escrito para personas no hablantes mediante visión artificial.

Abstract

The lack of a translation system is a barrier to communication and learning for non-speakers across various domains. Therefore, it is essential to develop and implement technological solutions. This paper proposes to design and implement a System for Translation from fingerspelling sign language to written natural language for non-speakers using computer vision. Both quantitative and qualitative approaches were employed in the methodology. The study used Mediapipe, a library that facilitates hand detection, tracking, and recognition from data structures to obtain information about 21 hand landmarks. The study population comprised students and teachers from the Specialized Deaf School of Chimborazo, and documentary and field research were conducted to gather information on the correct articulation of the Ecuadorian fingerspelling alphabet. The confusion data confirmed the system’s efficiency, achieving a performance of 100%. The system was evaluated under different lighting conditions, yielding 100% gesture detection in optimal conditions (over 1000 lux). Under moderate lighting conditions (100 to 900 lux), precision remains high at a 94% detection rate. Even under low-light conditions (less than 100 lux), the system maintains good detection capability with an 88% accuracy rate. Its high precision and reliability in gesture detection enable fluent and secure interaction across various contexts, from casual conversations to academic and professional environments.