Motion Recognition Based on Spatial Temporal HMM and Improved KPCA

In order to solve the problem that traditional intelligent surveillance is easily influenced by blocking and the capture views is limited, this paper presents a new method with 3 Kinects. Kinects are used to capture the human skeleton data and extract motion features. Principle component of raw data is extracted by using improved KPCA. Classifier is generated by using spatial-temporal Hidden Markov Model. A set of specific motions is analyzed in monitoring area. Experimental results show that this method can efficiently solve the problems that blocking and skeleton data is incomplete. It can also improve the recognition accuracy. The improved KPCA can improve the cumulative contribution rate and reduce the motion recognition time.