Improved OpenPose and Attention Mechanism for Sports Start Correction

The development of sports science and computer vision technology has made human pose recognition have great potential for application in sports. Considering the limitations of starting motion correction, the study proposes to use improved OpenPose posture estimation to achieve motion information ana...

Full description

Saved in:
Bibliographic Details
Published in:Informatica (Ljubljana) Vol. 48; no. 23; pp. 1 - 16
Main Author: Jiang, Bin
Format: Journal Article
Language:English
Published: Ljubljana Slovenian Society Informatika / Slovensko drustvo Informatika 01.12.2024
Subjects:
Accuracy
Algorithms
Athletes
Computer science
Computer vision
Correlation analysis
Data analysis
Deep learning
Efficiency
Human body
Human motion
Methods
Neural networks
Recognition
ISSN:0350-5596, 1854-3871
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The development of sports science and computer vision technology has made human pose recognition have great potential for application in sports. Considering the limitations of starting motion correction, the study proposes to use improved OpenPose posture estimation to achieve motion information analysis, and introduces attention mechanism to perform correlation analysis on athletes' limb posture information. Specifically, based on OpenPose's limitations in detecting subtle actions, inter frame information association and pose information classification processing are performed. Secondly, design a lightweight attention module and a network model under time attention mechanism for corrective analysis. The experimental results show that the improved algorithm has a starting posture recognition accuracy of over 0.85, and compared with the Dynamic Time Warping (DTW) model and the high-resolution human keypoint detection network (HRNet-w48), its recognition accuracy for the three stages of the starting action is over 93%. The ROC results indicate that the maximum accuracy of the research method's posture recognition can reach 0.97, which is superior to other baseline models, and its detection and recognition time for starting actions is less than 0.1 seconds. The methods proposed by the research institute have broad application prospects in improving sports performance and scientific training, and can provide suggestions and references for athletes' movement correction.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0350-5596
1854-3871
DOI:10.31449/inf.v48i23.6759