Application of Hidden Markov Model in Human Motion Recognition by Using Motion Capture Data

Abstract

The present study outlines an approach that describes how the statistical pattern recognition tool as Hidden Markov Model (HMM) is used to segment, extract features from different types of human common locomotion data, classify them and make a reliable probabilistic recognition. We have utilized six type of motion (climb, forward jump, jump, run, sit and walk), and each of them was composed of eight AMC data format specifically chosen from the CMU online library. The captured data were transformed from three dimensional (3D) joints trajectories into Matlab analyzable 2-Dimensional matrix using the Singular Value Decomposition (SVD) features factorization algorithm and then trained using the Baum-Welch algorithm, which has performed a classification of each motion’s extracted features for computational evaluation. The maximum iterations tolerance according to the number of motion was around 0.001 for highly accurate probability. In spite of the noisiness of the hand and toe trajectories, the recognition level was pretty higher than expected. Although, this investigation doesn’t take in account the different motion’s trajectories speed. The overall model has shown the probabilistic efficiency of HMM when it is about segmenting and evaluating vectorized motion’s features. The proposed study is suitable to any human common locomotion capture in AMC format and can be a great tool for further modeling in motion prediction.