SMARTFIGHT: análise de movimentos em vídeos de treinamento de muaythai com visão computacional

Abstract

This work proposes the development of a system for analyzing movements in Muay Thai training videos using Computer Vision and temporal analysis techniques. The solution employs human pose estimation algorithms to extract 2D joint coordinates from athlete performances using the YOLOv11-Pose model. These data are converted into CSV files and analyzed using methods such as the Fast Fourier Transform (FFT) to identify motion periodicity and Dynamic Time Warping (DTW) to assess similarity between executions. The system enables detection of strike cycles, rhythm analysis, and comparison with reference models, serving as a support tool for technical evaluation and training. Results demonstrate that the system can effectively segment and analyze strike executions, offering potential robust applications in combat sports and technology-assisted instruction.