AI-Based Recognition and Analysis of Gatka Martial Arts Techniques Using Human Pose Estimation and Machine Learning

Abstract

Abstract— Gatka is a traditional Sikh martial art, which is defined by synchronized movement of the body, the weapons, the defensive actions and the quick transitions. Teaching and evaluation are mostly based on the expertise of teachers and the process of observation, making it hard to quantify, provide quick feedback, and systematically preserve in digital format. This paper proposes a framework called AI-GATKA which utilizes machine learning and human pose estimation to achieve the aim of recognizing and analysing basic Gatka techniques. Media Pipe Pose processes video frames to extract body landmarks such as shoulder, elbow, wrist, hip, knee and ankle positions. Landmark points are transformed to posture, joint-angle, alignment and temporal-motion features. A Support Vector Machine (SVM), Random Forest and Neural Network classifiers are then employed to determine five technique categories: basic stance, forward strike, defensive block, side movement, and circular movement. The data set used in the prototype configuration reported in the source study consists of 250 samples, with 80% of the data used for training and 20% of the data used for testing. The SVM achieved 86%, the Random Forest 91% and the Neural Network 94% accuracy in the classification. This framework shows how the use of artificial intelligence with poses can help with technique recognition, training, objective assessments and digital preservation of Gatka. The paper also outlines the main drawbacks of camera viewpoint, lighting, among practitioners, occlusion, and fast movements of weapons. Main keywords—Artificial Intelligence, Computer vision, Gatka recognition, Human action recognition, Machine Learning, MediaPipe Pose, Pose-based Analysis.