[registration] redundant codes

[YeHuanjie]

I found redundant codes in ppf.h line96 and ppf_registration.hpp line153

Eigen::Vector3f model_point_transformed = transform_mg * model_point;
float angle = std::atan2 ( -model_point_transformed(2), model_point_transformed(1));
if (std::sin (angle) * model_point_transformed(2) < 0.0f)
    angle *= (-1);
p.alpha_m = -angle;

and

const Eigen::Vector3f scene_point_transformed = transform_sg * scene_point;
float alpha_s =
    std::atan2(-scene_point_transformed(2), scene_point_transformed(1));
if (std::sin(alpha_s) * scene_point_transformed(2) < 0.0f)
    alpha_s *= (-1);
alpha_s *= (-1);

There are too many (-1) in codes. In fact, they are redundant.

To verify my hypothesis, I conducted the following experiment.
I run the above codes and simplified code, comparing their results.

    for (int x = -10; x <= 10; x++)
    {
        for (int y = -10; y <= 10; y++)
        {
            float float_x = (float) x;
            float float_y = (float) y;
            // simplified
            float vanilla_alpha_s = std::atan2(float_y, float_x);
            // pcl
            float pcl_alpha_s = std::atan2(-float_y, float_x);
            if (std::sin(alpha_s) * float_y < 0.0f)
            {
                pcl_alpha_s *= (-1);
            }
            std::cout << "x: " << x << " y: " << y
                << " vanilla_alpha_s: " << vanilla_alpha_s
                << " magic_alpha_s: " << pcl_alpha_s  << std::endl;
        }
    }

So, the codes can be simplified as follows:

Eigen::Vector3f model_point_transformed = transform_mg * model_point;
float angle = std::atan2 (model_point_transformed(2), model_point_transformed(1));
p.alpha_m = -angle;

However, I am confused with the (-1). It makes no sense here.

[YeHuanjie]

According to 《A Method for 6D Pose Estimation of Free-Form Rigid Objects Using Point Pair Features on Range Data》, I guess the (-1) should be here:

[mvieth]

I am not sure if I completely understand what you suggest. Maybe create a pull request or post a git diff, then I can see the exact changes you propose?
Although, looking at the scatter plot, "simplified" and "pcl" differ at -pi and +pi? I am unsure whether this might cause any problems in the rest of the ppf code?

[YeHuanjie]

Apologies for the late reply, I just got back from vacation.
Here is the pull request:
#6230

Although, looking at the scatter plot, "simplified" and "pcl" differ at -pi and +pi? I am unsure whether this might cause any problems in the rest of the ppf code?

As far as I know, PI and -PI are the same for rotation axis.
And my experiments show that results are the same w/o this commit.