Pencak Silat Movement Classification Using CNN Based On Body Pose

Abstract

Pencak silat, besides from being useful for self-protection, also has many other benefits, such as increasing physical strength, maintaining posture, and maintaining heart health. Due to the recent pandemic, practicing pencak silat is difficult to do together. Even when there is study material on pencak silat at school, it is difficult for the sports teacher to teach the movements directly. Pencak silat exercises that are practiced alone without a coach can cause injury if the movements are not correct. Therefore, this study builds a system to recognize pencak silat movements. The system was built using the bodypose-based CNN method. Bodypose estimation is used to detect human body keypoints, then these keypoints are used as a feature for input to CNN to recognize movement in each frame. This system uses CNN because it requires fewer parameters and less computing power so that it can be more easily applied for further studies. The accuracy obtained reaches 77% when tested on data that has never been used. This model can be used as a starting point for creating an easy-to-use system to help people practice pencak silat with more recognizable moves.