CCD camera calibration is a necessary procedure to realize optical 3D profile measurement, its accuracy depends on the positioning accuracy of feature points in a great degree. A high accuracy corner detection method based on improved SV algorithm was proposed, through analyzing the deficiency of the existing methods to detecting black and white checkerboard corner points. Firstly, the corner points were detected in pixel level using the operators of symmetry and variance. Secondly, the simple gray bilinear interpolation was used for the area of a 5×5 window, whose center was the initial location corner in the photographed checker calibration image. Finally, centroid algorithm was adopted in the areas of each corner neighborhood after interpolation to obtain gray centroid coordinates. According to the amplification factor, centroid coordinates were located to sub-pixel corner coordinates corresponding to the calibration image. The experiment prove the validity of the proposed method, which locate sub-pixel corner coordinates , also can realize the high CCD calibration accuracy. The experiment results show that the average error of CCD calibration is 0.108 mm.