| dc.description.abstract | Pencak silat is a traditional Indonesian martial art that emphasizes precision in
technique, especially in kicking movements. Errors in kicking techniques can lead to
injuries, particularly in the leg muscles. Novice practitioners often struggle to
understand proper posture and technique, especially when training without instructor
supervision. This study proposes a technology-based solution that provides real-time
feedback to support independent training. The developed system utilizes the YOLOv11
algorithm to detect objects and estimate body pose in real time. The dataset was derived
from video extractions of front and side kick movements performed by practitioners
aged 11–17, resulting in a total of 3,619 images. The preprocessing stages included
cropping, keypoint annotation using MediaPipe, bounding box generation, resizing,
and data augmentation. The dataset was then split into training, validation, and testing
sets. The system successfully classified four categories of movement: correct/incorrect
front kick and correct/incorrect side kick. Evaluation results showed an accuracy of
92.5%, precision of 92.3%, recall of 92.1%, and an F1-score of 92.2%. The best
detection results were achieved in side kicks, while the most misclassifications occurred
in front kicks due to undetected foot soles. Although MediaPipe improved pose
estimation accuracy, it increased computational load, leading to delays on lowerspecification devices. Nevertheless, the system still provides real-time feedback in the
form of audible alarms and error notifications, helping novice practitioners improve
their techniques independently and reduce the risk of injury. This system can serve as
an effective training tool for learning basic pencak silat kicking techniques. | en_US |
