Muay Thai Posture Classification using Skeletal Data from Kinect and k-Nearest Neighbors

Muay Thai (also known as Thai boxing) is a martial art originated from Thailand. It is a discipline to make a barehanded fight by efficiently combining eight limbs. Because of its uniqueness, Muay Thai has become popular internationally. Many people travel to Thailand in order to study Muay Thai. At the same time, many Muay Thai schools have been opened in many countries. To efficiently utilize limbs or parts of the body in the combat, Muay Thai trainees need to study and practice various postures. Each posture is a combination of multiple limbs to make an attack. All the Muay Thai practices are based on one-on-one training. The trainees learn postures and practice them for a period of time. In this paper, we present an idea to apply Microsoft Kinect for Muay Thai practicing support. To achieve the goal, we propose a technique that identifies a Muay Thai posture from each body movement of player. Using Kinect, we can collect a sequence of skeletal data from each body movement. We then extract a number of features, and apply 1-NN technique based on Dynamic Time Warping to identify Muay Thai postures. In our experiment, we use straight punch, swing punch and upper cut postures to test our system.