| Sumario: | Muay Thai postures are the distinctive stances and movements used in traditional Thai boxing, concentrating on balance, strength, and fluid transitions between offensive and defensive techniques. Therefore, this study presents a novel system for detecting and categorizing Muay Thai postures through video processing and deep learning methods, which are skilled at examining the coordinated movements of two people fighting on the stage. The procedure involves identifying human formations and evaluating incorporated positions during duplicative movements, and organizing the postures operating a hybrid CNN and LSTM model. Moreover, to improve learning efficiency with joint position sequence data, the MinMaxScaler technique is used for data normalization. The type is based on an extensive dataset of 590 videos featuring various backgrounds, divided into 80% for training and 20% for testing. The resulting model completes an overall type accuracy of 84%, providing a strong basis for future applications in solely and paired of Muay Thai analysis. This approach offers potential advantages in sports science, competitor training, and motion analysis, allowing real-time posture recognition and performance evaluation. It helps coaches and practitioners identify incorrect techniques, reduce injury risks, and increase the accuracy of battle movements. In addition, future developments could include real-time integration into training applications, support for 3D pose analysis, and expanded datasets to improve the model's accuracy and validity across various conditions and practitioners.
|
|---|