Computer Vision Technology for Constructing a Swing Action Recognition and Feedback System

doi:10.4018/IJITSA.396710

Computer Vision Technology for Constructing a Swing Action Recognition and Feedback System

2026-01-01

0

PRE

AI

DOI：10.4018/IJITSA.396710

Original

Original request for help

Save

Abstract

En 中文

This study aims to construct a swing action recognition and feedback system that can be directly applied to table tennis teaching and training to realize real-time evaluation and targeted guidance of athletes' technical actions. Firstly, a dual-view high-speed acquisition scheme was designed. The red-green-blue and depth information were weighted and fused. Combined with dynamic threshold and motion detection, stable background segmentation was completed to ensure the accuracy of skeleton key point extraction. Secondly, a key frame extraction mechanism driven by the combination of speed peak and skeleton posture change rate was introduced before spatiotemporal modeling. Finally, the local joint and whole-body posture features were fused, and trajectory curvature parameters were introduced to describe the racket movement path. Combined with the convolution-attention hybrid network, action classification was realized.

Keywords:

Table Tennis Swing Action Recognition

Multi-Modal Visual Fusion

Trajectory Curvature Characteristics

Personalized Teaching Feedback

Computer Vision Technology for Constructing a Swing Action Recognition and Feedback System

Abstract

Journal

Organization