[filters] CropHull unexpected number of intersections

Context
I am using CropHull to test if a single point is inside the hull of a sphere. In applyFilter3D(PointCloud &output) function of crop_hull.hpp, I get unexpectedly high number of intersections for the given point.

Here is what I added to the function (using this fix) :

    bool isPointInsideHull = (crossings[0]&1) + (crossings[1]&1) + (crossings[2]&1) > 1;
    if ((crop_outside_ && isPointInsideHull) || (!crop_outside_ && !isPointInsideHull)) {
      cout << "crossing ray 0 : " << crossings[0] << endl;
      cout << "crossing ray 1 : " << crossings[1] << endl;
      cout << "crossing ray 2 : " << crossings[2] << endl;

      cout << "point is inside hull !-------------------------------------------------------------- " << endl;
      this_thread::sleep_for(10000ms);
      output.push_back (input_->points[(*indices_)[index]]);
    }

Current Behavior
This is the resulting number of intersections I got for each ray :

crossing ray 0 : 1405                                                                                                   
crossing ray 1 : 1407                                                                                                   
crossing ray 2 : 1397 

Expected behavior
I expected crossings array to contain either 0 (no intersection, possibility that the point is outside), 1 (1 intersection, possibility that the point is inside) or 2 values (2 intersections, possibility that the point is outside of the sphere).

To Reproduce
I use PCL 1.8.0 on Windows 10. Here is the link to the files used :

• polygon.ply contains the hull information
• unit_test.ply is the input point cloud we want to test

@taketwo @kunaltyagi

Start in crop_hull.hpp, specifically applyFilter3D(PointCloud &output), and reproduce the PCL 1.8.0 Windows 10 case with polygon.ply and unit_test.ply. Compare the crossings calculation with the expected 0–2 intersections and inspect the referenced fix for context. Done when the cause of the roughly 1,400 intersections is identified and the reported point-in-hull behavior matches expectations.