Fix issue 2817

[DumDereDum]

Pull Request Readiness Checklist

resolves #2817

There was an error in the work with NaNs in depth.

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 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]

wants to merge 7 commits

7 commits are too much for single line fix.
Please keep 1 commit in PR history. Do not over-complicate review.

As stated in OpenCV contribution guidelines all "fixup" commits must be removed from the history (squashed).

[savuor]

Patching NaNs with 0 is incorrect, it leads to false surfaces in the frame (parts of a plane z=0) instead of absent data.
What should be done is to keep NaNs at their places after applying bilateral filtering.
(The true solution here is to implement KinFu preprocessing pipeline the way the odometry is implemented, i.e. with valid pixels mask, but this involves all KinFu rewriting).

[savuor]

As it turned out, computePointsNormals ignores zero values, so it's OK to patch depth with NaNs as soon as it's done in internal copy.