Recognizing the brazilian signs language alphabet with neural networks over visual 3D data sensor

Abstract

This paper describes an investigation of the Brazilian Signs Language (LInguagem BRAsileira de Sinais - LIBRAS) alphabet recognition using neural networks. The LIBRAS alphabet is represented by static postures and dynamic gestures. In this investigation, gestures were recorded with a 3D camera sensor and its associated library, providing coordinates from hands' fingertips. Deaf people, LIBRAS teachers and students were involved in the recording process. The pre-processing data involved frames sampling, normalization and 3D geometric transformations. Neural network models with different settings were trained, compared, and assessed to verify classification accuracy.