使用OpenCv 3.0校准FishEye镜头 [英] FishEye lens calibration with OpenCv 3.0

问题描述

我必须用鱼眼镜头校准我的相机.

I have to calibrate my camera with a Fisheye lens.

首先，我将OpenCv 2.4与CV_CALIB_RATIONAL_MODEL标志一起使用，以获得更好的结果，但这还不够. 之后，我发现OpenCv 3.0具有用于校准的鱼眼功能.

First I used OpenCv 2.4 with the flag CV_CALIB_RATIONAL_MODEL to get better results but it wasn't enough. After, I have seen that OpenCv 3.0 have fisheye function for calibration.

我计算我的objectPoints并使用findChessboardCorners()获得imagePoints，它们似乎是正确的.

I compute my objectPoints and get my imagePoints with findChessboardCorners(), they seem correct.

我的问题是使用 fisheye :: calibrate . 我返回的值，例如重投影误差，本征矩阵和失真系数是错误的.它们是-nan值.

My problem is after, with fisheye::calibrate. My returned values such as reprojection error, intrinsic matrix and distorted coefficients are wrong. They are -nan values.

有人知道我在做什么错吗?

Does anyone have any idea what I'm doing wrong?

谢谢

推荐答案

对于新的OpenCV 3.0鱼眼函数，我发现除非指定cv::fisheye::CALIB_RECOMPUTE_EXTRINSIC标志(例如38像素重投影错误)，否则我将无法获得可用的固有校准.另外，我指定了cv::fisheye::CALIB_FIX_SKEW.当然，您应该指定尽可能多的cv::fisheye::CALIB_FIX_Kx标志.

For new OpenCV 3.0 fisheye functions, I found I did not get a usable intrinsic calibration unless cv::fisheye::CALIB_RECOMPUTE_EXTRINSIC flag was specified (e.g. 38 pixel reprojection error). Also, I was specifying cv::fisheye::CALIB_FIX_SKEW. Of course, you should specify as many cv::fisheye::CALIB_FIX_Kx flags as you can.

对于我的镜头，这是一个非常简单的微型镜头，但径向变形较大，正常的OpenCV校正功能无法产生准确的校正.通过上面的标志并修复了K2=K3=K4=0，我能够获得良好的结果.

For my lens, which is a very simple, miniature lens but with big radial distortion, the normal OpenCV rectification functions did not yield accurate rectifications. I was able to get good results with the above flags and fixing K2=K3=K4=0.

这篇关于使用OpenCv 3.0校准FishEye镜头的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！