Reconhecimento de gestos em vídeos utilizando modelos ocultos de Markov e redes neurais convolucionais aplicado a LIBRAS

This work proposes a system for recognition of a sequence of dynamic gestures in videos using a combination of convolutional neural networks to model the hand shapes and hidden Markov models to model the gestures. A total of 15 signs from the Brazilian Sign Language are used. The accuracy achieved reached 100% under certain restrictions, such as a single gesticulator, the use of gloves and a controlled environment.