Inverse kinematics and gesture pattern recognition using Hidden Markov Model on BeatMe! project: Traditional dance digitalization

摘要

Indonesian traditional dance preservation efforts that nowadays increasingly eroded by foreign culture needs to be improved and adapted to technological improvement. To answer that, BeatMe! Project developed for fulfil the needs of entertainment and traditional dance learning media by integrate 3D motion capture, processing data and visualization. This paper will explain the processing data detail implementation which include big data summarizing as in summarizing XYZ coordinate of joint to be angle between two joints. In addition, this paper also explain detail implementation of HMM learning system for gesture pattern learning and recognizing. Environment used is Kinect, Visual Studio and MATLAB. Result show summarized data of discrete value between joints angle, learning curve as the learning process output that tends to rise and converge on a value within limits of 700.000 symbols, new gesture pattern recognition show a good performance in one degree joint because its position nearest the main torso, and not to good performance in two degree joint because of randomized value that happen when the body position isn't aligned with Kinect.