Autonomous recognition of martial arts forms using RGB-D cameras

The work presented here proposes an approach to automate the process of detecting and evaluating martial arts forms. This document presents a detection system on a set of static poses, which will be expanded in the future into full-movement forms. The results show that it is feasible to build a Kinect-based stance recognition system using simple machine learning algorithms, which will serve as a basis for more complex work in the future.

I. Introduction

In a martial arts context, a form is a well-defined set of motions and stances which are used to teach students the techniques and movements of a discipline. A form may include: hands and legs attacks, throwing, wrestling and many other actions (like the use of weapons), which were developed by the Masters of the particular style.