Fit a partial point cloud with a superquadric

Solve an optimization problem to find out the best superquadric that fits a given partial point cloud.

Note

The superquadric is parametrized in terms of its center, shape, the principal axes, and a rotation angle around the z-axis, in order to account for objects that are supposed to be lying on a table parallel to the x-y plane (to keep things simple 😉).

:warning: If the input point cloud does not guarantee that the above assumptions hold, it will need to be first transformed through a convenient rototraslation.

The equation of the superquadric is the following:

$\left( \left| \frac{x-c_x}{s_x} \right|^\frac{2}{\epsilon_2} + \left| \frac{y-c_y}{s_y} \right|^\frac{2}{\epsilon_2} \right)^\frac{\epsilon_2}{\epsilon_1} + \left| \frac{z-c_z}{s_z} \right|^\frac{2}{\epsilon_1} = 1$

Dependencies

• YARP
• iCub
• Ipopt
• VTK

Command-line options

• --file file-name: specify the file containing the point cloud given in the following plain format:

  x0 y0 z0 [r0 g0 b0]
  x1 y1 z1 [r1 g1 b1]
  ...

  RGB colors are optional.

• --remove-outliers "(<radius> <minpts>)": outliers removal based on spatial density clustering. The aggregation of points in clusters is regulated through the distance radius, whereas minpts represents the minimum number of points of a valid cluster. Only points belonging to the largest cluster will survive as inliers.
• --uniform-sample <int>: specify the integer step for performing uniform down-sampling as follows:
  • 1 means no down-sampling
  • > 1 enables down-sampling
• --random-sample <double>: specify the percentage in [0,1] for performing random down-sampling.
• --inside-penalty <double>: specify how much to penalize points that will lie inside the superquadric's isosurface wrt points lying outside (default = 1.0).
• --disable-viewer: specify not to launch the viewer.
• --color "(<r> <g> <b>)": change the color of the superquadric by specifying RGB components as double in the range [0,1].
• --opacity <double>: specify the opacity of the superquadric as double in the range [0,1].
• --background-color "(<r> <g> <b>)": change the background color by specifying RGB components as double in the range [0,1].

Real-time mode

If no --file option is passed through the command line, the module will open up a port called /find-superquadric/points:rpc to which the point cloud can be sent as a yarp::sig::PointCloud<yarp::sig::DataXYZRGBA> object. Then, the module will reply with the superquadric parameters:

center-x center-y center-z angle size-x size-y size-z epsilon-1 epsilon-2

The angle around the z-axis is returned in degrees, whereas center-* and size-* are expressed in the same length units of the input point cloud.

Example

$ find-superquadric --remove-outliers "(0.01 10)" --random-sample 0.2 --file ./data/cylinder

Output

👨🏻‍💻 Maintainers

This repository is maintained by:

	@pattacini