基于GPU的大视场景物畸变实时校正算法

Real-time calibration algorithm of distortion of wide-field view based on GPU

针对使用短焦距镜头获取大视场景物信息时,利用光学方法校正光学畸变难度大、成本高的问题,提出了一种利用数字图像处理技术对光学系统进行畸变校正的新算法。该算法在采用CUDA技术的可编程图形处理器的基础上,利用标准网格模型来求解包括径向、离心以及薄棱镜在内的多项镜头畸变系数,大大提高了校正精度和计算速度,从而达到高质量实时校正的目的。通过相应实验验证了该算法的可行性和高效性,针对1920*1080像素的高清视频校正速度可达30帧/s,校正精度达亚像素级。

Using short focal length lens to get the information of wide-field view always leads to image distortion due the existence of optics distortion. Calibration with optic method is demonstrated quite difficult and expensive. To solve these problems, this paper proposes a new algorithm to calibrate the distortion of optic system using digital image processing technique, called GPU-Based Grid Calibration （GBGC） algorithm. On the bases of programmable Graphics Processing Unit （GPU）, which adopts CUDA technology, this algorithm solves several lens distortion coefficients, such as radial distortion, centrifugal distortion and thin-prism distortion and so on, by standard grid model. This algorithm can maximize the calibration precision and calibration speed, moreover, accomplish high quality real-time calibration. Experiment results demonstrate the feasibility and efficiency of this algorithm, and the calibration precision extends to sub-pixels.