Badminton Action Classification Based on Human Skeleton Data Extracted by AlphaPose

In the realm of sports analytics, the precise classification of intricate movements is paramount. Badminton, a sport celebrated worldwide, is a testament to this with its swift and complex actions. While the domain of human action recognition has seen significant advancements, the specific exploration of badminton action classification using skeleton keypoint data remains a largely uncharted territory. This study bridges this research gap. Utilizing the AlphaPose framework, we harnessed skeleton keypoint data from selected badminton videos, offering a novel approach to understanding and categorizing the sport's nuanced movements. Our analysis showcased the superiority of the Long Short-Term Memory (LSTM) model, achieving an impressive 80% prediction accuracy, in contrast to the Convolutional Neural Network (CNN) model's 60%. This distinction emphasizes LSTM's adeptness at handling sequential data inherent to skeleton keypoints. In essence, this research not only underscores the untapped potential of skeleton keypoints in revolutionizing badminton analytics but also paves the way for future endeavors in this specialized domain.