Allows structured_light pipeline to be run from Python

[r2d3]

SinusoidalPattern::unwrapPhaseMap now takes an InputArray instead of InputArrayOfArrays to correct a Python binding problem

present a scriptable HistogramPhaseUnwrapping::create

This patch solves issue #1704 and allow to write a whole pipeline including reliability computation.

Pull Request Readiness Checklist

See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request

• I agree to contribute to the project under OpenCV (BSD) License.
• To the best of my knowledge, the proposed patch is not based on a code under GPL or other license that is incompatible with OpenCV
• The PR is proposed to proper branch
• There is reference to original bug report and related work
• There is accuracy test, performance test and test data in opencv_extra repository, if applicable
  Patch to opencv_extra has the same branch name.
• The feature is well documented and sample code can be built with the project CMake

[alalek]

Thank you for contribution!

It would be nice if you could add simple test for Python API (no need to perform complex accuracy check).

See example: https://github.com/opencv/opencv_contrib/blob/4.2.0/modules/rgbd/misc/python/test/test_rgbd.py
or similar Python tests from main opencv repository.

Test data is stored in opencv_extra repository.

[alalek]

Please follow the existed practice and avoid unpacking of "Params" structure.
You can check this code as an example:

https://github.com/opencv/opencv/blob/850414a501d5e6dba111c92d73fdd05794c7061d/modules/features2d/include/opencv2/features2d.hpp#L653-L654

[r2d3]

The existing practice is not clear as the structured_light itself is using 2 practices:

• Params unpacking in GrayCodePattern::create
• smart pointer for SinusoidalPattern::create

Some codes pass a smart pointer like this is done for SinusoidalPattern::create

https://github.com/opencv/opencv_contrib/blob/master/modules/structured_light/include/opencv2/structured_light/sinusoidalpattern.hpp#L100

but I followed the solution used by GrayCodePattern::create

https://github.com/opencv/opencv_contrib/blob/master/modules/structured_light/include/opencv2/structured_light/graycodepattern.hpp#L89

Because changing the create API (instead of overloading it) will break current codes using structured_light.

The code you point out is not using Params struct to pass parameters.

If you just pass a const Params&, the Python binding generator will not generate the needed code. Am I right ?

[alalek]

"Params" class should be marked for bindings wrapping too (use CV_EXPORTS_W_SIMPLE)

[r2d3]

OK. I managed to bind the HistogramPhaseUnwrapping::Params; What I was missing is the typedef in pyopencv_phase_unwrapping.hpp as it is done in pyopencv_features2d.hpp

I added the Python test for structured_light replicating what is done in C++. However, this shown a "bug" in HistogramPhaseUnwrapping::unwrapPhaseMap where the unwrappedPhaseMap were uninitialized in places outside of the mask.
It is not clear if this were intentional or not.

[alalek]

Well done! Thank you for contribution 👍

[hitmanman]

Hi the following codings make my python kernel die.
The problem is
'unwrapped_phase_map = sinus.unwrapPhaseMap(wrapped_phase_map, cam_size, shadow_mask)'
Can you tell me how to correct it?
thank you

import cv2
import numpy as np
import math
import sys

cam_size = [-1,-1]

params = cv2.structured_light_SinusoidalPattern_Params()
params.width = 1024 #pixel width
params.height = 1024 #pixel height
params.nbrOfPeriods = 20 # amount of waves
params.setMarkers = False # patterns /w or /wo markers
params.horizontal = False
params.methodId = cv2.structured_light.FTP
#params.methodId = cv2.structured_light.PSP
#params.methodId = cv2.structured_light.FAPS
params.shiftValue = 2.0 * math.pi / 3.0
params.nbrOfPixelsBetweenMarkers = 70

sinus = cv2.structured_light_SinusoidalPattern.create(params)
patterns = sinus.generate()
count = 0

img = []
while count < len(patterns[1]) :
img.append(patterns[1][count])
count += 1

captures = []
captures.append(img[0])
captures.append(img[1])
captures.append(img[2])

wrapped_phase_map, shadow_mask = sinus.computePhaseMap(captures)
cam_size = (img[0].shape[0], img[0].shape[1])
print("before")
unwrapped_phase_map=np.zeros([img[0].shape[0], img[0].shape[1]],np.uint8)
unwrapped_phase_map = sinus.unwrapPhaseMap(wrapped_phase_map, cam_size, shadow_mask)
print("after")

[r2d3]

Hi the following codings make my python kernel die.

The problem is
'unwrapped_phase_map = sinus.unwrapPhaseMap(wrapped_phase_map, cam_size, shadow_mask)'
Can you tell me how to correct it?

Which version of OpenCV are you using ?

import cv2
print(cv2.__version__)

You should have a recent 3.4 release (ideally 3.4.10) to get the correction

Regards