我想将单应性应用于以下几点:
array([[-7.4894, 1.8873],
[-7.4973, 1.8543],
[-7.5375, 1.6725],
[-7.5681, 1.522 ],
[-7.5961, 1.371 ],
[-7.6252, 1.2013],
[-7.6504, 1.031 ],
[-7.667 , 0.8985],
[-7.6817, 0.7657],
[-7.6954, 0.613 ],
[-7.7054, 0.4786],
[-7.7124, 0.3452],
[-7.7182, 0.1931],
[-7.7215, 0.0866],
[-7.7716, 0.0872],
[-7.7715, 0.0929],
[-7.7651, 0.2884],
[-7.7587, 0.4269],
[-7.7528, 0.5233],
[-7.7418, 0.6616],
[-7.7275, 0.8116],
[-7.7048, 1.0032],
[-7.6916, 1.0988],
[-7.6686, 1.2478],
[-7.6352, 1.4379],
[-7.6091, 1.5741],
[-7.5784, 1.7219],
[-7.538 , 1.8995],
[-7.4894, 1.8873]], dtype=float32)
我的相机单应性矩阵是这样的:
array([[ 3.9643041e-04, 6.5913662e-07, 3.1965813e-03],
[ 7.4297395e-07, -3.9652368e-04, -4.4492882e-04],
[-9.3076696e-06, -3.5773560e-06, 1.0000000e+00]], dtype=float32)
当我尝试使用cv2.perspectiveTransform应用单应性时,出现以下错误:
`cv2.error: OpenCV(4.0.0) C:\projects\opencv-python\opencv\modules\core\src\matmul.cpp:2270: error: (-215:Assertion failed) scn + 1 == m.cols in function 'cv::perspectiveTransform'`
我怀疑每个点都需要另一个维度。但是我不确定如何用numpy添加此维度。
解决此问题的正确方法是什么,是否有确定根本原因的方法?该错误消息对我而言意义不大。
答案 0 :(得分:1)
确保pts形状为(n, 1, 2)
或(1,n,2)
:
pts = np.float32(pts).reshape(-1,1,2)
#pts = np.array([pts], np.float32)
cv2.perspectiveTransform(pts, M)
例如:
pts = np.array([[1,2,],[3,4]], np.float32)
M = np.array([[ 3.9643041e-04, 6.5913662e-07, 3.1965813e-03],
[ 7.4297395e-07, -3.9652368e-04, -4.4492882e-04],
[-9.3076696e-06, -3.5773560e-06, 1.0000000e+00]], dtype=np.float32)
## (n, 1, 2)
pts1 = pts.reshape(-1,1,2).astype(np.float32)
dst1 = cv2.perspectiveTransform(pts1, M)
## (1, n, 2)
pts2 = np.array([pts], np.float32)
dst2 = cv2.perspectiveTransform(pts2, M)
结果:
>>> print(pts1)
[[[1. 2.]]
[[3. 4.]]]
>>> print(dst1)
[[[ 0.00359439 -0.00123725]]
[[ 0.00438869 -0.00202888]]]
>>> print(pts2)
[[[1. 2.]
[3. 4.]]]
>>> print(dst2)
[[[ 0.00359439 -0.00123725]
[ 0.00438869 -0.00202888]]]
这是另一个示例:
How do I use the relationships between Flann matches to determine a sensible homography?