Machine Learning Approach for Ibing Penca Stance Recognition Using Landmark Detection and Angle-Based Classification

Supporting independent learning of traditional martial arts presents a challenge when direct instructor guidance is unavailable. This study explores the feasibility of recognizing Ibing Penca stances using an interpretable, computer vision–based pose representation. The proposed system aims to support independent practice by identifying and classifying 62 Ibing Penca stances under controlled conditions. Image and video data were collected using an Orbbec camera and processed through a pose landmark detection pipeline. Human body landmarks were extracted using MediaPipe, which provides 33 keypoints representing major joints and body segments.Based on these landmarks, six joint angles corresponding to the right arm, left arm, right leg, left leg, right foot, and left foot were computed using three-point angle calculations. These angle features were then used in a rule-based classification framework to represent stance configurations. Experimental results indicate that the proposed angle-based representation can distinguish most stance prototypes within the constructed dataset, achieving a stance-level recognition rate of 91% (57 out of 62 stances) under controlled conditions. Rather than claiming generalizable performance, this study is positioned as an initialfeasibility investigation. Future work will focus on expanding the dataset, incorporating greater movement variability, and evaluating performance undermore diverse environmental and subject conditions.