تصحیح انحراف لنزهای عریض
Abstract
1. Introduction
2.Distortion Calibration
3.Distortion correction
4.Results and analysis
5.Conclusion
References
Abstract
An adaptive piecewise low-order mapping method that automatically extracts control point pairs from single orthogonal grid pattern is proposed to correct serious wide-lens distortion. Firstly, distorted control points (DCPs) are located by oriented searching along crossing lines in thinned grid image, and the relative distribution relationship between DCPs is obtained to estimate ideal control points in normalized grid pattern. Subsequently, a piecewise polynomial optimization problem, which minimizes two cost functions, is built to model whole radial distortion, thus adaptive pieces and corresponding segmented coefficients are optimized. Finally, an inverse mapping table is established to improve correction efficiency. Experimental results illustrate that the proposed method is effective and practical for high resolution wide-angle lens with perfect precision, and completely preserves the undistorted image edge.
Introduction
Compared with images taken by ordinary cameras, images from wide-angle lens contain more information in a wider range of view owing to shorter focal length, larger view and deeper field. Therefore, wide-angle cameras have been widely used in video surveillance, semi-automatic parking, auxiliary driving, endoscopic imaging and other computer vision fields [1-2]. However, due to nonlinear geometric imaging structure, images captured by wide-angle cameras suffer from serious barrel distortion [3] including decentering distortion, thin prism distortion and radial distortion [4]. Radial distortion is the typical distortion that displaces pixels points towards the principal intersection point that lens optical axis passes through the image[5], which twists the original image geometry. Only such distortion is corrected, can wide-angle lens really apply to practical requirements [6]. To overcome this issue, an increasing number of research works have been carried out. In Tsai’s two-step method [7], a radial distortion model was presented and corresponding points known in 3D space were utilized to recover distortion parameters. Based on Tsai’s method, Junhee Park et al [8] proposed a distortion model defined on ideal undistorted coordinates to achieve fast and simple lens mapping.