Action Recognition Method of Basketball Training Based on Big Data Technolog

Abstract

Aiming at the problem that improper posture of basketball players leads to not obvious sports effects, the present paper proposes an action recognition method combining computer vision and big data technology and applies it to athletes' daily training and competition. Firstly, based on the current mainstream motion recognition models, 3D graph convolution are used to improve the original 3D convolution to promote the expression ability of spatial structure features and temporal features in skeleton sequences. Secondly, channel and spatial attention mechanisms are introduced to focus on the weight distribution of key points and strong features in different posture recognition processes. Finally, the proposed model is tested in real data, and the test results show that the model runs smoothly while maintaining high recognition performance. It can more effectively direct basketball players to implement comprehensive, systematic, and scientific teaching and training standards that directly support raising the game's general level of performance.