In order to solve the problem that the calibration of fisheye camera with automatic parking vehicle is time-consuming, an improved chessboard corner detection method based on growth is proposed in this paper. On the basis of the original algorithm, the proposed hybrid distortion model can quickly determine the two initial directions of corners in different regions, so as to quickly improve the speed of rough positioning. At the same time, according to the particularity of production line calibration, the threshold of double standard deviation is proposed as the criterion to judge the initial chessboard. The experimental results in the production line environment show that the algorithm can not only meet the accuracy requirements of calibration, but also greatly improve the detection speed, and has a certain practical value for the calibration of production line camera.