Because physical camera parameters do not exist independently in equation but couple to each other, the camera parameters obtained by existing calibration methods do not actually match their physical meanings. Decomposing parameter directly from projection matrix can cause considerable error. To cope with this problem, a stepwise method for camera calibration is proposed. Since pose estimation is sensitive to equivalent focal length, equivalent focal length can be calibrated based on pose estimation variance; then principal point is calibrated by homographic constraint; extrinsic camera parameters are calibrated by orthogonal iteration algorithm; finally the distortion coefficients are calculated. Experiments have proved this method′s correctness and high precision, and show that this method can effectively avoid coupling of camera parameters, in which way each parameter can match its own physical meaning pretty well, and can be used independently.