|
The fisheye lens, with a field of view (FOV) angle reaching 180 degrees, is highly effective for visual inspection and measurement of large scenes, provided it is accurately modeled and precisely calibrated. Unlike perspective projection lenses with narrower FOVs, fisheye lenses exhibit stronger nonlinearity, greater distortions, and more pronounced aberrations. Moreover, the calibration accuracy of current state-of-the-art imaging models and methods is generally less than 1/10 of a pixel, which fails to meet the high-precision requirements of certain measurement scenarios. In this paper, we introduce a new imaging model and calibration method for fisheye cameras. Supplementing traditional projection and distortion models, we propose image-variant intrinsic orientation parameters to account for the influence of the camera's attitude on intrinsic parameters. Additionally, we develop a corresponding bundle adjustment algorithm for this model. Because traditional calibration objects are too small to meet the high-precision needs of fisheye lenses, we establish a large planar calibration field with numerous control points and capture a series of images from various orientations for bundle adjustment calibration of the imaging model. To address the significant impact of initial parameter values on the bundle adjustment convergence process, we present a new calibration method that enables automatic processing and matching of calibration image data, ensuring robust and reliable results. Calibration experiments using a NIKON D810 camera and a Nikkor 16mm fisheye lens demonstrate that our method achieves a calibration precision of 1/15 of a pixel, surpassing other models and methods reported in the literature. Furthermore, our proposed method is distinguished by its simplicity in operation and automated data processing.
|
