As an important part of vehicle assisted driving, Bird Eye View (BEV) can not only effectively improve driving safety, but also participate in the control decision-making of unmanned driving as a data source of ground information. Therefore, it is necessary to generate a distortionless BEV with complete ground information. In this paper, through in-depth study of the imaging and de-distortion characteristics of fisheye cameras, iterative calibration and FOV overlapping area weighted fusion methods are introduced into the existing methods to solve the existing BEV generation problems, such as loss of ground information and imperfect matching of Field of View (FOV) overlapping areas, due to the special distortion characteristics of vehicle-mounted fisheye cameras and actual installation restrictions. The experimental results show that the improved BEV generation method has higher computational efficiency and image quality.