Abstract: In initialization stage of markers in a motion capture system,it needs to establish the correspondence between each marker and joint,which causes a fussy and heavy workload because it is usually carried out manually.To solve this problem,we proposed a reliable automatic marker initialization method.Firstly,we predefine the actor's initial posture.Secondly a PCA algorithm is performed to analyze the body's two principal axes.According to the definition of human skeleton model,the correspondences are established between markers and joints.And then the 3D positions of markers with joint information are reconstructed.Finally other viewing points can be initialized by projecting 3D position to each camera plane.Several experimental results demonstrate the proposed method's accuracy and efficiency.