Feature barcode detector parameters

[cristidbr-adapta]

Attempt to solve #24902 without changing the default detector behaviour.
Megre with extra: opencv/opencv_extra#1150

Introduces new parameters and methods to cv::barcode::BarcodeDetector.

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 Apache 2 License.
• To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
• The PR is proposed to the proper branch
• There is a reference to the 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

[asmorkalov]

@mshabunin could you take a look?

[cristidbr-adapta]

Just saw #23889 and I had verified that this PR is unable to solve the orientation delta issue. However, I believe users should be able to tune at least some subset of the parameters of bardetect.cpp.

Benefits and use-cases:

• localisation for high resolution inputs Make BarcodeDetector resolution invariant #24902
• pair the opencv detector with alt decoder (native mobile devices have resolution invariant decoders, e.g. iOS)

@mshabunin this can happen without breaking userland, but I don't like the idea of pushing additional tests wrt the algo state. You seem to have a better understanding of how it works.
@asmorkalov the ArUco params impl looks elegant. Is that the standard way users would expect to adjust such values for classes in opencv::objdetect?

[mshabunin]

I believe extracting parameters to a separate structure is not necessary in this case, it can be done using get/set methods (see comments by @opencv-alalek).

Also I'm thinking whether passing window sizes array should be done using InputArray/OutputArray style, maybe it could be simple vector<double>?

We will also need new test case or even several. Perhaps existting test images can be resized or augmented to illustrate properties effect.

Are there any alternative names for the new properties? In my opinion they are too long and complex, for example word "detector" is not techically necessary. windowSizes are actually scales. Piece thresh usually should placed at the end - setDownSampleThresh instead of setThreshDownSample. setGradientThresh?

New parameter documentation should include estimated effect on quality, performance, detection count, etc. Maybe some of parameters better work when they correlate with other parameter or image dimensions? Allowed values range should be documented and verified in the set method.

[cristidbr-adapta]

Thanks for the detailed feedback, it is clear how to proceed now. Will have some time to address this over the next couple of days.

[asmorkalov]

@cristidbr-adapta Friendly reminder.

[cristidbr-adapta]

@mshabunin added set/get methods for all parameters and vector<double> for scales. Removed unnecessary parameter for NMS box threshold as I was able to derive that based on the impl internals.

Regarding tests, I still need to tweak it for adding a regression to handle the _detector_sm.jpg file in extra. Allowed ranges are not asserted yet. Impact to performance should be expected when the default behaviour is changed to allow larger input sizes.

Docs TBD.

[cristidbr-adapta]

For docs, should I update the existing tutorial with new methods and bump version to 4.10?

[asmorkalov]

👍

[asmorkalov]

@mshabunin @opencv-alalek could you take a look again.

[asmorkalov]

@mshabunin @opencv-alalek Friendly reminder.

[asmorkalov]

@dkurt Could you take a look?

[mshabunin]

Can we avoid creating extra test images?

I suggest the following test scenario:

1. take one of existing images
2. set detector parameters to some predefined values
3. run detector and verify that result is as expected (e.g. it can be empty with some parameter set)

Test can be parametrised with (params set, expected result). Test can be combined with new regression test (check parameters after setting).

Input image can be modified in some way (scaled, rotated, filtered,...) or it can be built from several other images (vconcat).

[cristidbr-adapta]

Can we avoid creating extra test images?

Extra test data was designed to detect abnormal changes to the barcode_detector.cpp algorithm impl wrt to false positives. It needs to happen.

[mshabunin]

What is different in these images comparing to pre-existing ones? If we really need them, can we reduce their size, for example we could add only one of them (_sm) and get crops from it in the test to replace other sizes or maybe somehow extend _lg image instead.

[asmorkalov]

👍

• Renamed public methods according suggestion.
• Use float for scales.
• Minor cleanup.