In order to improve the calibration accuracy of camera parameters, the improved Zernike moment and eccentricity error correction were used to locate the center of the circle with high accuracy for the problem of blurred edges and eccentricity of spatial circular projection when calibrating with circular feature points in stereo vision system. The problem of blurred edges of circular feature projection image caused by uneven illumination intensity at the calibration site of stereo vision imaging system was firstly considered, and the Gaussian error function was introduced to describe the grayscale distribution of the edge transition section. The Gaussian edge model was established, and Zernike moments of the projection image were calculated based on this model, and then the improved Zernike moments were used to achieve high-precision circular feature projection edge pixel coordinate positioning. In addition, the factors that affecting the deviation between the projection point of the circular feature and the fitted circle center were analyzed, and the deviation compensation of the iterative fitted circle center was made based on this analysis to make it close to the real circle center projection, and lastly the circle centers of 99 circular marker points were extracted and used for the calibration of the camera parameters by the proposed algorithm. The simulation experiments show that the accuracy of the algorithm for edge positioning of the projected image and the accuracy of the circle center fitting are higher than those of the traditional algorithm; In the practical measurement experiments, it turns out that the length measurement accuracy of standard rod based on the proposed method is improved by 30% compared with traditional methods.