CODEWITHSUNDEEP
pythonopencvnumpy
user1688482
user1688482

Reputation: 73

OpenCV estimateAffine3D failes with cryptic error message

I have 2 3D point clouds, stored in numpy.ndarrays, containing either 0 or 1 (Indication of a detected point or none). I would like to compute the affine transformation that converts the 'Test' array to the 'Reference' array.

I call the function like this:

import numpy as np
from cv2 import estimateAffine3D

[...] Read in the Arrays [...]

print(np.shape(Reference))
print(np.shape(Test))

AffineTransfMatrix = estimateAffine3D(Reference,Test)

And the Output I get is:

(132, 2055, 701)
(132, 2055, 701)
OpenCV Error: Assertion failed (count >= 0 && to.checkVector(3) == count) in estimateAffine3D, file /home/wenzlern/libraries/opencv/modules/calib3d/src/ptsetreg.cpp, line 513
Traceback (most recent call last):
  File "/home/wenzlern/code/python/AbsorbtionSpecAnalysis/AlignEnergies.py", line 67, in <module>
    estimateAffine3D(Reference,Test)
cv2.error: /home/wenzlern/libraries/opencv/modules/calib3d/src/ptsetreg.cpp:513: error: (-215) count >= 0 && to.checkVector(3) == count in function estimateAffine3D

I have tried playing around with the data type, using Reference/Test.astype('float32'), but could not change the result. The documentation does not seem to specify a specific format.(http://docs.opencv.org/2.4/modules/calib3d/doc/camera_calibration_and_3d_reconstruction.html#estimateaffine3d).

Does anyone have an idea what could go wrong? Am I missing something, or calling the function wrong?

Thanks a lot, Nils

Upvotes: 3

Views: 3908

Answers (2)

Hendrik D
Hendrik D

Reputation: 71

I've found that, with OpenCV 4.1.1, handing a numpy array with dtype that isn't np.int64 will also give a similar error message for estimateAffinePartial2D and estimateAffine2D:
error: (-215:Assertion failed) count >= 0 && to.checkVector(2) == count in function 'estimateAffinePartial2D'

Upvotes: 2

user1688482
user1688482

Reputation: 73

User Micka (in the comments to the question) had the right solution. The Array has to be passed in Nx3 format. In other words: The coordinates (x,y,z) of every point in the Point-Cloud are one 1x3 entry in the Nx3 matrix to be passed.

Thanks!

Upvotes: 1

Related Questions