Normal Estimation Using Integral Images not always working

Your Environment

• Operating System and version: Ubuntu 18.04
• Compiler: GCC
• PCL Version: 1.9.1

Context

I'm trying to estimate normals of a brick with sizes 20x20x30cm. I have following patch and was wondering if there is maybe also a bug lurking in the integral normal estimation, since as I see it this PR just fixes the other normal estimation method.

Expected Behavior

The normals of one side of the brick get estimated pretty accuretaly, but completely fail on the other side. I suspect that this is due to the other side being more sparse, but it does not look so bad that I would excpect it to deliver no normals at all.

Current Behavior

Code to Reproduce

Pretty much the sample code from the tutorial:

ne_.setNormalEstimationMethod (ne_.COVARIANCE_MATRIX);
ne.setNormalSmoothingSize(normal_smoothing_size);
ne.setInputCloud(input_cloud);
ne.compute(*output_cloud);

Possible Solution

I tried different smoothing sizes in increments of 5 starting at 5 to 40, but no size delivers any normals for the top plane. Is this to be expected or a bug? Attached is a test file.

Start with the normal estimation using integral images tutorial and its sample code, then reproduce the behavior with the attached test file on PCL 1.9.1 under Ubuntu 18.04 and GCC. Compare smoothing sizes from 5 through 40 and verify whether normals are produced for the sparse top plane as well as the other brick surfaces.