Python 校准单个摄像机时，RMS值较低，而在stereoCalibrate中，RMS值较高

我正在尝试校准一个由两个相同的

佳能EOS60D

组成的立体设置，我将使用它进行摄影测量。 为了校准摄像机，我使用了高精度的

circlegrid

模板。我能够使用cv2.findCirclesGrid检测圆的中心点，并且能够很好地校准单个摄像头。当我尝试使用OpenCV中的

cv2.stereoCalibrate

功能校准两个摄像头时，问题就出现了。 我将阐明所采取的步骤，并发布一些代码片段和我得到的输出。如果有人能指出我做错了什么，我将不胜感激

步骤 1） 第一步是使用OpenCV中的

cv2.findCircleGrid

函数检测circlegrid模式

2） 然后使用从每个摄像头检测到的网格模式，我使用cv2.CalibleCamera来获得摄像头矩阵和两个摄像头的失真系数

3） 在获得两台摄像机的摄像机矩阵和失真系数后，我将它们传递给

cv2.stereoCalibrate

函数

输出

校准左摄像机后的输出如下

Left Camera Return Value :  0.202690712694
Left Camera Camera Matrix :  [[  2.42046647e+04   0.00000000e+00       1.78281995e+03]
[  0.00000000e+00   2.42115121e+04   2.51720578e+03]
[  0.00000000e+00   0.00000000e+00   1.00000000e+00]]
Left Camera Dist Coeff :  [[-0.09448288 -1.38934436  0.00951455 -0.00568115  0.11248399]]

Right Camera Return value :  0.258429588138
Right Camera Camera Matrix :  [[  2.69094574e+04   0.00000000e+00   1.70580157e+03]
[  0.00000000e+00   2.69534209e+04   2.55185400e+03]
[  0.00000000e+00   0.00000000e+00   1.00000000e+00]]
Right Camera Dist Coeff :  [[-0.1854594   1.89124255  0.00592814 -0.00377697  0.05715002]]

StereoCalib Return Value :  278.98818985
Rotation Vector :  [[ 0.72415334  0.13289823  0.67671265]
[ 0.68078781  0.018949   -0.73223554]
[-0.11013584  0.99094853 -0.07675351]]
Translation Vector  [[-1.20860215]
[ 1.23017549]
[ 2.08116679]]
Essential Matrix :  [[-1.55231939  1.17960457  1.42948401]
[ 1.37397347  1.47424592  1.31558743]
[-1.7136373  -0.18638995  0.05250614]]
Fundamental Matrix :  [[ -3.15554184e-07   2.39721211e-07   6.99264445e-03]
[  2.78844610e-07   2.99110045e-07   5.21248054e-03]
[ -9.54712002e-03  -2.19149124e-03   1.00000000e+00]]

校准右摄像机后的输出如下

Left Camera Return Value :  0.202690712694
Left Camera Camera Matrix :  [[  2.42046647e+04   0.00000000e+00       1.78281995e+03]
[  0.00000000e+00   2.42115121e+04   2.51720578e+03]
[  0.00000000e+00   0.00000000e+00   1.00000000e+00]]
Left Camera Dist Coeff :  [[-0.09448288 -1.38934436  0.00951455 -0.00568115  0.11248399]]

Right Camera Return value :  0.258429588138
Right Camera Camera Matrix :  [[  2.69094574e+04   0.00000000e+00   1.70580157e+03]
[  0.00000000e+00   2.69534209e+04   2.55185400e+03]
[  0.00000000e+00   0.00000000e+00   1.00000000e+00]]
Right Camera Dist Coeff :  [[-0.1854594   1.89124255  0.00592814 -0.00377697  0.05715002]]

StereoCalib Return Value :  278.98818985
Rotation Vector :  [[ 0.72415334  0.13289823  0.67671265]
[ 0.68078781  0.018949   -0.73223554]
[-0.11013584  0.99094853 -0.07675351]]
Translation Vector  [[-1.20860215]
[ 1.23017549]
[ 2.08116679]]
Essential Matrix :  [[-1.55231939  1.17960457  1.42948401]
[ 1.37397347  1.47424592  1.31558743]
[-1.7136373  -0.18638995  0.05250614]]
Fundamental Matrix :  [[ -3.15554184e-07   2.39721211e-07   6.99264445e-03]
[  2.78844610e-07   2.99110045e-07   5.21248054e-03]
[ -9.54712002e-03  -2.19149124e-03   1.00000000e+00]]

从上面的输出可以看出，

cv2.calibleCamera

的两个返回值都是合理的

校准立体声设置后的输出如下

Left Camera Return Value :  0.202690712694
Left Camera Camera Matrix :  [[  2.42046647e+04   0.00000000e+00       1.78281995e+03]
[  0.00000000e+00   2.42115121e+04   2.51720578e+03]
[  0.00000000e+00   0.00000000e+00   1.00000000e+00]]
Left Camera Dist Coeff :  [[-0.09448288 -1.38934436  0.00951455 -0.00568115  0.11248399]]

Right Camera Return value :  0.258429588138
Right Camera Camera Matrix :  [[  2.69094574e+04   0.00000000e+00   1.70580157e+03]
[  0.00000000e+00   2.69534209e+04   2.55185400e+03]
[  0.00000000e+00   0.00000000e+00   1.00000000e+00]]
Right Camera Dist Coeff :  [[-0.1854594   1.89124255  0.00592814 -0.00377697  0.05715002]]

StereoCalib Return Value :  278.98818985
Rotation Vector :  [[ 0.72415334  0.13289823  0.67671265]
[ 0.68078781  0.018949   -0.73223554]
[-0.11013584  0.99094853 -0.07675351]]
Translation Vector  [[-1.20860215]
[ 1.23017549]
[ 2.08116679]]
Essential Matrix :  [[-1.55231939  1.17960457  1.42948401]
[ 1.37397347  1.47424592  1.31558743]
[-1.7136373  -0.18638995  0.05250614]]
Fundamental Matrix :  [[ -3.15554184e-07   2.39721211e-07   6.99264445e-03]
[  2.78844610e-07   2.99110045e-07   5.21248054e-03]
[ -9.54712002e-03  -2.19149124e-03   1.00000000e+00]]

但立体声校准的返回值太高

参数

我在cv2.stereoCalibrate中使用的标志和标准如下

Left Camera Return Value :  0.202690712694
Left Camera Camera Matrix :  [[  2.42046647e+04   0.00000000e+00       1.78281995e+03]
[  0.00000000e+00   2.42115121e+04   2.51720578e+03]
[  0.00000000e+00   0.00000000e+00   1.00000000e+00]]
Left Camera Dist Coeff :  [[-0.09448288 -1.38934436  0.00951455 -0.00568115  0.11248399]]

Right Camera Return value :  0.258429588138
Right Camera Camera Matrix :  [[  2.69094574e+04   0.00000000e+00   1.70580157e+03]
[  0.00000000e+00   2.69534209e+04   2.55185400e+03]
[  0.00000000e+00   0.00000000e+00   1.00000000e+00]]
Right Camera Dist Coeff :  [[-0.1854594   1.89124255  0.00592814 -0.00377697  0.05715002]]

StereoCalib Return Value :  278.98818985
Rotation Vector :  [[ 0.72415334  0.13289823  0.67671265]
[ 0.68078781  0.018949   -0.73223554]
[-0.11013584  0.99094853 -0.07675351]]
Translation Vector  [[-1.20860215]
[ 1.23017549]
[ 2.08116679]]
Essential Matrix :  [[-1.55231939  1.17960457  1.42948401]
[ 1.37397347  1.47424592  1.31558743]
[-1.7136373  -0.18638995  0.05250614]]
Fundamental Matrix :  [[ -3.15554184e-07   2.39721211e-07   6.99264445e-03]
[  2.78844610e-07   2.99110045e-07   5.21248054e-03]
[ -9.54712002e-03  -2.19149124e-03   1.00000000e+00]]

flags=cv2.CALIB\u FIX\u固有

criteria=cv2.TERM\u criteria\u MAX\u ITER+cv2.TERM\u criteria\u EPS，1000，1e-6

如果有人能指出我哪里做错了，我将不胜感激。
Hi@the_parzival我也有同样的问题。你解决了这个问题吗？@MikelDíezBuil我没办法解决，但我可以通过改变帧之间的旋转来减少它。在进行校准时，您可以遵循一些良好的规则。