3D Motion Recognition based on Ensemble Learning

In this paper, a novel method is presented for 3D motion recognition based on motion capture database. We use 3D features and their key spaces of each human joint to represent human motion. After features extraction, ensemble HMM learners are used to train data. Then each action class is learned with one HMM and bagging algorithm is used to ensemble all learners. Since ensemble learning can effectively enhance supervised learners, ensembles of weak HMM learners are built. It is obvious that the proposed methods are effective by experimental results.