Human motion analysis in an intelligence surveillance system is a hot research topic in computer vision

and temporal segmentation of human activity sequence is the most fundamental step in human motion analysis.In this paper

an unsupervised online temporal segmentation algorithm is presented

and then the segmentation result is recognized by HMM.Firstly

a robust shape encoding scheme is employed to produce a compact representation of human silhouette

and a new feature called contour point set is proposed.Secondly

the intrinsic dimensionality of feature sequence and the corresponding low-dimensional manifolds are determined using SVD

and the break of projecting error of activity sequence on the determinate manifolds is detected as the segmentation point of the activity sequence.Temporal segmentation results are recognized by HMM finally.Experiments on two public databases show the effectiveness of the segmentation and recognition algorithms in this paper.