Article: Stereo vision and deep learning-based movement correction for sports dance Journal: International Journal of Information and Communication Technology (IJICT) 2024 Vol.25 No.9 pp.60 - 75 Abstract: In sports dance training, dancers' wrong movements are unavoidable, and if they are not corrected in time, it will not only reduce the effect of dance expression, but also directly related to the improvement of sports performance. Therefore, we suggest a correction method for sports dance movements based on stereo vision and deep learning. Firstly, a binocular stereo imaging model is established by using the principle of triangle similarity. Secondly, 3D CNN is used to extract spatio-temporal features from the preprocessed images, and the early attention mechanism is introduced to adaptively enhance the key features that are beneficial to early action prediction. Finally, the important features are used to model the action boundaries by estimating the relative probability distribution of the action boundaries to obtain the recognition results. Simulation experiments show that the accuracy and peak signal-to-noise ratio are 91.17% and 20.45 dB, respectively. Inderscience Publishers - linking academia, business and industry through research

Title: Stereo vision and deep learning-based movement correction for sports dance

Authors: Wumei Jiang; Hui Lan

Addresses: School of Physical Education, JiMei University, Xiamen 361000, China ' Chengyi University College, JiMei University, Xiamen 361021, China

Abstract: In sports dance training, dancers' wrong movements are unavoidable, and if they are not corrected in time, it will not only reduce the effect of dance expression, but also directly related to the improvement of sports performance. Therefore, we suggest a correction method for sports dance movements based on stereo vision and deep learning. Firstly, a binocular stereo imaging model is established by using the principle of triangle similarity. Secondly, 3D CNN is used to extract spatio-temporal features from the preprocessed images, and the early attention mechanism is introduced to adaptively enhance the key features that are beneficial to early action prediction. Finally, the important features are used to model the action boundaries by estimating the relative probability distribution of the action boundaries to obtain the recognition results. Simulation experiments show that the accuracy and peak signal-to-noise ratio are 91.17% and 20.45 dB, respectively.

Keywords: stereo vision; deep learning; motion correction; 3D CNN; attention mechanism.

DOI: 10.1504/IJICT.2024.143329

International Journal of Information and Communication Technology, 2024 Vol.25 No.9, pp.60 - 75

Received: 27 Aug 2024
Accepted: 13 Sep 2024
Published online: 13 Dec 2024 *

Title: Stereo vision and deep learning-based movement correction for sports dance

Keep up-to-date