Abstract: This paper addresses the problem about interior knowledge understanding of 3D human model.It proposes a new algorithm to detect 3D human landmarks and get hierarchical segmentation result by using semantic knowledge.Firstly,the algorithm formulates a domain knowledge representation by using 3D human landmarks and their parametric definition.Secondly,based on domain knowledge and geometry attribute,the algorithm automatically recognizes semantic landmarks of the input human model.In addition,the hierarchical segmentation result of the input model can be easily obtained by using detected semantic landmarks and geodesic distance.Finally,graph cut is used to refine coarse segmentation result and the shape index is used to improve the detection accuracy of landmarks.In experiment,the proposed algorithm shows very robust under human pose variation.Furthermore,it can avoid the inconsistent segmentation problem,which often occurs in the existing algorithms.