Abstract: A new approach to calibrate camera parameters by image pair is presented. With this method, the parameters of a camera is calculated by using the invariance of lines under the projective transformation, different from previous methods which estimate the fundamental matrix by the image characteristics. Two parallel lines on each of the image pair were extracted. The focus and the extrinsic parameters were solved according to the projective invariance of parallel lines with the center of the image taken as initialization of the principal point. Based on the polar constraint, the nonlinear optimization was employed to compute the precise parameters of the camera. The algorithm needs less information of the scene and accurate self-calibration is easily realized. The recovery of 3D scene structure can be accomplished with the new algorithm. Experimental results on the synthetic and real images demonstrate the feasibility of this algorithm.