Question

我是立体声校准的新手并尝试获取立体相机的矩阵然后重新映射图像，因此我可以将它们用于生成深度图。我使用了示例图片left*.jpg和right*.jpg以及此代码的部分内容：http://docs.opencv.org/3.0-beta/doc/py_tutorials/py_calib3d/py_calibration/py_calibration.html#calibration（以及来自samples/python/calibrate.py）

所以我的代码如下：

import cv2
...
img_names = [glob('../data/left*.jpg'), glob('../data/right*.jpg')]
obj_points = [[], []]
img_points = [[], []]
h = ... #
w = ... #

... # get the obj_points and img_points from both image sets like in the examples
# This works fine, I viewed the results with cv2.drawChessboardCorners

rms_l, camera_matrix_l, dist_coeffs_l, _, _ = cv2.calibrateCamera(obj_points[0], img_points[0], (w, h), None, None)
rms_r, camera_matrix_r, dist_coeffs_r, _, _ = cv2.calibrateCamera(obj_points[1], img_points[1], (w, h), None, None)

stereo_flags = 0
stereo_flags |= cv2.CALIB_USE_INTRINSIC_GUESS
stereo_flags |= cv2.CALIB_FIX_INTRINSIC

# run stereo calibration
rms_stereo, camera_matrix_l, dist_coeffs_l, camera_matrix_r, dist_coeffs_r, R, T, E, F  = cv2.stereoCalibrate(obj_points[0], img_points[0], img_points[1], camera_matrix_l, dist_coeffs_l, camera_matrix_r, dist_coeffs_r, (w, h), flags=stereo_flags)

# run stereo rectification
rectification_matrix_l, rectification_matrix_r, projection_matrix_l, projection_matrix_r, _, _, _ = cv2.stereoRectify(camera_matrix_l, dist_coeffs_l, camera_matrix_r, dist_coeffs_r, (w, h), R, T)

map1, map2 = [], []
map1[0], map2[0] = cv2.initUndistortRectifyMap(camera_matrix_l, dist_coeffs_l, rectification_matrix_l, projection_matrix_l, (w, h), cv2.CV_16SC2)
map1[1], map2[1] = cv2.initUndistortRectifyMap(camera_matrix_r, dist_coeffs_r, rectification_matrix_r, projection_matrix_r, (w, h), cv2.CV_16SC2)

print ... # see below

for in in range(2):
    img = cv2.imread(img_names[i][0], 0)
    imgt = cv2.remap(img, map1[i], map2[i], cv2.INTER_LINEAR)
    cv2.imshow('image', imgt)
    cv2.waitKey(0)

但是，我的结果图像非常奇怪（左右）：

现在是我的问题，我的错误是什么？我是否计算了矩阵不正确或插入了错误的参数？我看到很多例子以类似的方式使用这些函数，但是他们的结果并不像我的那样。

编辑：以下是我的矩阵：

---------- Camera Stereo ----------
RMS: 54.0798238093
camera matrix left:
[[ 532.80990768    0.          342.49522241]
 [   0.          532.93344826  233.88792572]
 [   0.            0.            1.        ]]
distortion coefficients left:  [ -2.81325825e-01   2.91151900e-02   1.21234424e-03  -1.40823847e-04
   1.54861062e-01]
camera matrix right:
[[ 537.43775429    0.          327.6113408 ]
 [   0.          536.95118843  248.89561922]
 [   0.            0.            1.        ]]
distortion coefficients right:  [-0.29741745  0.14930176 -0.00077008  0.00032599 -0.06565751]
R:
[[ 0.81520177  0.24703546  0.52385071]
 [ 0.42427129  0.36098612 -0.83047149]
 [-0.39425873  0.89925664  0.18946647]]
Rectification matrix Left :
[[ 0.84817656  0.5225405  -0.08687895]
 [-0.44783099  0.61975759 -0.64447493]
 [-0.28292036  0.58553561  0.75967369]]
Rectification matrix Right :
[[ 0.77500946  0.62063732  0.11903635]
 [-0.54957916  0.56894011  0.61177602]
 [ 0.31196648 -0.5395521   0.7820233 ]]
Projection Matrix Left :
[[ 448.32924881    0.          328.40793514    0.        ]
 [   0.          448.32924881  215.26811713    0.        ]
 [   0.            0.            1.            0.        ]]
Projection Matrix Right :
[[  4.48329249e+02   0.00000000e+00   3.28407935e+02   0.00000000e+00]
 [  0.00000000e+00   4.48329249e+02   2.15268117e+02   1.97345858e+04]
 [  0.00000000e+00   0.00000000e+00   1.00000000e+00   0.00000000e+00]]

opencv cv2.remap创建奇怪的图像

0 个答案: