Considering the fact that the camera calibration method based on perspective projection model yield intricate and unstable result, a new calibration algorithm based on affine approximation was presented. The affine projection matrix was led out and then some parameters were decomposed. Under the presumption of taking image center as principal point, other parameters were solved. At last, all the parameters' precision was improved on the basis of the cost function of re-projection error. Simulation and real image experiments show that the new method can produce more stable result than the usual one.