Aguilar J. 2D-to-3D Lifting of Sign Language Body Poses with Recurrent Neural Networks. Giró-i-Nieto X, Duarte A. Barcelona: UPC ETSETB TelecomBCN; 2021 .  (340.44 KB)

Abstract

This paper aims at improving the quality of a dataset that contains multiple sequences of 3D poses extracted from American Sign Language videos. Each pose consists of 147 points with three coordinates each. We propose an algorithm able to correct missing points as well as to add some constraints such as the length of the bones. To prove the quality of the algorithm’s outcome, we evaluate the task of lifting 2D to 3D poses with a deep learning model trained on raw data, and another one trained with the preprocessed data.

  

Projects

Sign Language Recognition, Translation and Production