On an approach focused on skeleton-based action recognition along with deep learning methodologies, this study aims to present the application of classification models in order to enable real-time assessment of Taekwondo athletes. For that was used a developed dataset of some Taekwondo movements, as data for Long Short-term Memory (LSTM), Convolutional Long Short-term Memory (ConvLSTM) and Convolutional Neural Network Long Short-term Memory (CNN LSTM) training, validation, and inference. The results obtained allow to conclude that for the system application defined as goals the data structure the LSTM model achieved the best results. The obtained accuracy value of 0,9910 states the model reliability to be applied in the system proposed. Regarding time response the LSTM model also obtained the best result with 288 ms.
Deep Learning in Taekwondo Techniques Recognition System: A Preliminary Approach
Lect.Notes Mechanical Engineering
International Conference Innovation in Engineering ; 2022 ; Minho, Portugal June 28, 2022 - June 30, 2022
2022-06-21
12 pages
Aufsatz/Kapitel (Buch)
Elektronische Ressource
Englisch
BASE | 2018
|ROTATION MOVEMENT ANLYSIS IN TAEKWONDO POWER BREAKING MOVEMENT OF 360° JUMP BACK KICK
BASE | 2013
|Traffic Sign Detection and Recognition Using Deep Learning Approach
Springer Verlag | 2023
|