When traditional Tsai two-stage technique is used to calibrate a camera, its calibration accuracy will be limited by first-order radial distortion model. Therefore, a camera imaging model considering both radial distortion and tangential distortion is proposed, and a new method to solve the model is presented to improve the distortion correction accuracy. As the center of an image has a smaller distortion, the image center points are used to give a linear equation group and to calculate a part of parameters of the camera. Then camera parameters are taken into comprehensive distortion model to derive the initial values of distortion coefficients. As focal length and matrix's translation value can not be calculated directly when the calibration board's depth variation is small, the values have to be retaken to model and to be calculated.Then the converged solution is approached by two-stage iterative computation. Finally, the world coordinate system is converted by space geometric transform, perspective transform and imaging transform to obtain the pixel coordinate of the reprojective image. The calibration error is obtained by calculating the mean difference between re-projection pixel coordinates and actual measured pixel coordinates of each point. Experimental result indicates that the calibration method can offer the accuracy of 0.1149 pixel, which is better than Tsai's 0.3670 pixel with well repeatability.