How Fast Is Sign Language? A Reevaluation of the Kinematic Bandwidth Using Motion Capture

Autor: Félix Bigand, Prigent, E., Berret, B., Annelies Braffort
Přispěvatelé: Bigand, Félix, Laboratoire Interdisciplinaire des Sciences du Numérique (LISN), CentraleSupélec-Université Paris-Saclay-Centre National de la Recherche Scientifique (CNRS), Centre National de la Recherche Scientifique (CNRS), Complexité, Innovation, Activités Motrices et Sportives (CIAMS), Université d'Orléans (UO)-Université Paris-Saclay, Institut Universitaire de France (IUF), Ministère de l'Education nationale, de l’Enseignement supérieur et de la Recherche (M.E.N.E.S.R.), Institut National de Recherche en Informatique et en Automatique (Inria)-CentraleSupélec-Université Paris-Saclay-Centre National de la Recherche Scientifique (CNRS)
Jazyk: angličtina
Rok vydání: 2021
Předmět:
Zdroj: 29th European Signal Processing Conference (EUSIPCO 2021)
29th European Signal Processing Conference (EUSIPCO 2021), 2021, Online streaming, France
HAL
Popis: International audience; Human motion lies within a range of low frequencies. Filtered and down-sampled motion capture (mocap) data can thus provide meaningful representations for computational models. However, little is known about the kinematic bandwidth of Sign Language (SL), apart from isolated signs. Studies examining isolated signs suggested that SL could be limited to relatively low frequencies. This is unlikely to be appropriate for real-life conditions where signs are produced faster and are combined with several other rapid motion features. The present study investigated the spectral content of a multi-signer mocap dataset of continuous signing in French Sign Language. Across six different signers, Power Spectral Density estimation and residual analysis of the mocap data revealed that SL motion can be limited to a 0-12-Hz bandwidth, which is substantially wider than stateof-the-art estimates on isolated signs. More specifically, filtering the movements below 6 Hz caused distortion of the rapid motion, which suggests that SL motion involves higher frequencies in real-life conditions. The importance of kinematic bandwidth estimation is further addressed with a machine learning model trained to identify the six signers of the dataset. The performance of the model significantly decreased when using inappropriate bandwidths.
Databáze: OpenAIRE