如何对3维运动数据进行自动识别，是提高数据利用效率和进行计算机动画创作的前提。为了有效地识别运动数据，需要把运动数据投影到非线性流型低维子空间中，先识别出运动的内在结构，然后对运动的各个关节点分别进行学习，最后基于集成学习的方法产生强的隐马尔可夫学习器，以便能够对一些常见的运动类型进行自动识别。实验结果表明，这种基于子空间集成的人体运动识别方法较传统方法的检索精度、检索速度均有较大提高。

In this paper, a motion retrieval and recognition system is investigated from a ensemble learning model. In order to recognize and retrieve 3D motion data, first motion features are extracted from motion data. Due to the high dimensionality of motion’s features, a generalized isomap nonlinear dimension reduction based on the estimation of underlying eigenfunction is used for training data of ensemble HMM learning. Then each action class is learned with one HMM. Since ensemble learning can effectively enhance supervised learning, ensembles of weak HMM learners are built. Experimental results show that our approaches are effective for information retrieval from large scale motion database.