The severe radial distortion of ultra-wide field of view (FOV) fisheye camera results in poor model fitting and challenges in calibration board detection. In this paper, a novel calibration method for ultra-wide FOV fisheye cameras is proposed based on improved camera model and SE(3) image pre-correction. Initially, a method to extend the maximum fitting FOV of the camera model to over 180 degrees is proposed. Subsequently, a calibration board detection approach is proposed using SE(3) image pre-correction. Specifically, image pre-correction is incorporated into the camera calibration process, utilizing SE(3) to define the pre-correction plane. Calibration boards are detected within the pre-corrected images, enhancing the reliability, accuracy and speed of board detection in distorted images, consequently increasing the control point's maximum FOV. Lastly, the improved camera model and SE(3) image pre-correction are integrated into a feedback-based camera calibration system for ultra-wide FOV fisheye cameras. Operating with real-time or offline video streams as input, this system autonomously selects calibration key frames, optimizes camera parameters and calibration board poses in real-time. Simulation and real-world experiments verify the effectiveness of the proposed method, leading to a 62% increase in the achievable maximum FOV.
