Utilization of Body Posture Feature Extraction with Movenet for Silat Movement Classification Using Artificial Neural Network (ANN)

Pencak Silat is a distinctive traditional martial art of Indonesia that needs to be preserved. One way to preserve Pencak Silat through a technological approach is by introducing its movements using Human Pose Estimation. Human Pose Estimation is a branch of computer vision that predicts the position of keypoints on a human body from an image, in this case, images of Pencak Silat movements. One of the models for extracting keypoints is MoveNet, developed by Google, excelling in striking a balance between speed and precision. This study proposes a classification system for Pencak Silat movements using the Artificial Neural Network (ANN) architecture. It utilizes a dataset of Pencak Silat movements with keypoints extracted using MoveNet. The dataset includes 8 movements: ‘dengkulan’, ‘pukulan lurus’, ‘pukulan samping’, ‘sikap pasang’, ‘sikutan atas’, ‘sikutan depan’, ‘tendangan A’, and ‘tendangan T’. The research employs K-Fold Cross Validation with K=5 for validation. The research results demonstrate that by utilizing the body posture feature extraction with MoveNet, it is possible to classify Pencak Silat movements using the Artificial Neural Network (ANN) method, achieving an accuracy of 0.971, which indicates a good performance.