Research of unsupervised posture modeling and action recognition based on spatial-temporal interesting points
RIS ID
38631
Abstract
Posture modeling is critical for action description and recognition,a posture modeling and action recognition method is proposed in this paper.Spatial Temporal Interesting Points (STIPs) are extracted from learning samples,in fact,one posture consists of a set of STIPs;a unsupervised clustering method is adopted to classify salient postures from these posture samples,then a GMM model is established for each clustering result;transitional probability among salient postures are calculated,and a Visible state Markov Model(VMM) is learnt to describe various actions.Bi-gram method is put forward for action recognition,Extensive experiments are conducted and the results prove its robustness and validity.
COinS
Publication Details
Wang, C., Liu, Y. & Li, W. (2011). Research of unsupervised posture modeling and action recognition based on spatial-temporal interesting points. Chinese Journal of Electronics, 39 (8), 1751-1756.