MediaPipe and Its Suitability for Sign Language Recognition
Identifikátory výsledku
Kód výsledku v IS VaVaI
<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F00216305%3A26210%2F23%3APU148801" target="_blank" >RIV/00216305:26210/23:PU148801 - isvavai.cz</a>
Výsledek na webu
<a href="https://www.engmech.cz/improc/2023/251.pdf" target="_blank" >https://www.engmech.cz/improc/2023/251.pdf</a>
DOI - Digital Object Identifier
—
Alternativní jazyky
Jazyk výsledku
angličtina
Název v původním jazyce
MediaPipe and Its Suitability for Sign Language Recognition
Popis výsledku v původním jazyce
The paper presents the framework MediaPipe as a tool to extract pose features for the task of word-level isolated sign language recognition. It tests the framework’s suitability on the state-of-the-art sign language dataset AUTSL. Extracted sequences of pose features are classified by the Long Short-Term Memory recurrent neural network constructed with the TensorFlow computational library. The paper describes the proposed method flow, preprocessing of the extracted features, and training. Obtained results are then validated on test datasets, and the impact of face landmarks is evaluated. The top-1 accuracy with face landmarks is 49.89 %, while 53.21 % without them.
Název v anglickém jazyce
MediaPipe and Its Suitability for Sign Language Recognition
Popis výsledku anglicky
The paper presents the framework MediaPipe as a tool to extract pose features for the task of word-level isolated sign language recognition. It tests the framework’s suitability on the state-of-the-art sign language dataset AUTSL. Extracted sequences of pose features are classified by the Long Short-Term Memory recurrent neural network constructed with the TensorFlow computational library. The paper describes the proposed method flow, preprocessing of the extracted features, and training. Obtained results are then validated on test datasets, and the impact of face landmarks is evaluated. The top-1 accuracy with face landmarks is 49.89 %, while 53.21 % without them.
Klasifikace
Druh
O - Ostatní výsledky
CEP obor
—
OECD FORD obor
10201 - Computer sciences, information science, bioinformathics (hardware development to be 2.2, social aspect to be 5.8)
Návaznosti výsledku
Projekt
—
Návaznosti
S - Specificky vyzkum na vysokych skolach
Ostatní
Rok uplatnění
2023
Kód důvěrnosti údajů
S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů