Efficient unsupervised behavioral segmentation for human motion capture data base on Gaussian mixture model
Xiaomin Yu, Weibin Liu, Weiwei Xing · 2016
In order to better reuse of motion capture data, complex motion sequences should be segmented into distinct behaviors. As we move toward collecting longer motion sequences, automatic behavior segmentation techniques are becoming important. In this paper, we proposed a method for automated segmentation motion capture data into distinct behaviors. We employ Gaussian Mixture Model (GMM) to model the entire sequence and segment sequences whenever two consecutive sets of frames belong to different Gaussian distribution. In order to avoid falling into local optimum, we use split Expectation Maximization (EM) algorithm to estimate parameters of GMM. we build an energy function to deal with sub-sequences noise. Extensive experiments are conducted on sequences performed by subject 86 of the CMU database, each of which is an association of roughly several natural actions (e.g. walking, running, punching, kicking). Results of experiments demonstrate that our proposed method is possible and robust.