Applying the example code gom.api.imaging.compute_point_from_pixels() results in an error

[resident12]

Applying the example code of gom.api.imaging.compute_point_from_pixels() to lookup the 3D coordinates of a 2D point alas results in an error. Would appreciate to be able to utilize the described functionality as stated in the documentation. Or should the data be pre-processed somehow to make it work? Please advise me.

#see https://zeissiqs.github.io/zeiss-inspect-addon-api/2023/python_api/python_api.html#gom-api-imaging-compute-point-from-pixels

import gom

DIRNAME = 'c:/data/'
MEASUREMENT_SERIES = 'Scan 1'
MEASUREMENT = 'M1'

gom.script.sys.close_project() #Start by closing all current projects
gom.script.sys.load_project(file= DIRNAME + 'ATOSQ12_1.ginspect')
gom.script.view.toggle_image_mapping(enable=True, use_animation=False) #Show it to me in the UI

measurement = gom.app.project.measurement_series[MEASUREMENT_SERIES].measurements[MEASUREMENT]

left = gom.api.project.get_image_acquisition(measurement, 'left camera', [0])[0]

p = gom.api.imaging.compute_point_from_pixels([[(gom.Vec2d (10, 10), left)]], False)

print(p)

Error in line 18

GomApiError: Computation of 2D points is not allowed in 3D projects.

Stacktrace:

Line 18 in 'gom-api-imaging-compute-point-from-pixels.py' ()

[mprinkezs]

Please try it with two (i.e. left and right) camera images as described in https://zeissiqs.github.io/zeiss-inspect-addon-api/2023/python_examples/point_pixel_transformations.html.
There are cases where a single image is sufficient, I am currently investigating the background.

[resident12]

Thank you for the reply. Yes, I can confirm it does work with two images. But, then the relation between left and right image-pixels must be known. We need the single image solution. A big thank you in advance. Your help is very much appreciated.

[mprinkezs]

The topic is not sorted out completely yet. What I learned now, is that a single image works only with 2D projects in ZEISS INSPECT Correlate. In this case, the measuring object must have a flat surface and the camera axis must be orthogonal to this surface.

[mprinkezs]

Unfortunately, a preceding step is required in this case. You use the function gom.api.imaging.compute_epipolar_line as a helper. You supply a set of tuples (pixel, image) and an additional image and get an epipolar line for the second image for each tuple. The wanted pixel-match is located on this epipolar line. (see https://en.wikipedia.org/wiki/Epipolar_geometry)
As the final step, you have to search for the wanted pixel on the epipolar line. This is typically done by gray value comparison of the surrounding pixels. As a prerequisite, the pixels must be located at distinct position, i.e. not on a uniformly colored surface. In the best case, the surface has been prepared with a speckle pattern.

[resident12]

Thank you. Please hold on. I'm still chewing on this one :-). Yes, the Epipolar-line is an option, but then the pixel matching must be redone.

Perhaps it can be done easier. For, in the Zeiss Inspect GUI I can switch on/off the visual overlay of the 2D-image, 3D-mesh and Reference-Points of a saved project => the sensor calibration matrix is stored with the data and is reread with the data from disk. I could apply that calibration matrix to solve the issue. But somehow I can't use the export via File->Export->Sensor Calibration. To be used under specific conditions only?

[mprinkezs]

May I ask you to refer to the ZEISS Quality Forum? This way, more people would be able to see and hopefully answer your question.

[mprinkezs]

Closing due lack of activity.