Shape descriptors for tabletops

In my Master's Thesis for the TICMA Master at the UPF I reviewed the shape descriptors and implemented three that can be used for control in tabletop systems, such as the reactable.


Here you can watch two videos showing the result of these descriptors:


reacTIVision is a computer vision intended to be used for tabletop systems.

A tabletop or surface computer is a flat surface similar to a table, that in fact is a big screen with multitouch capabilites. Is easier to understand what it is by looking for example at Microsoft Surface which is a commercial tabletop system.

reacTIVision is an open source vision system that can be used for the development of tabletop systems. Typically, a vision-based tabletop system has a camera under the semitransparent surface of the table which allows the computer to understand what is on the table. The information captured with the camera is the input for the tabletop (just as the keyboard and mouse is the input for a normal computer), and then the tabletop projects images in its surface (just like the screen of a computer).

The camera that is under the table "sees" the objects that are on the table, reacTIVision analyzes the image and sends the information to the application that projects images to the surface

reacTIVision can identify the fiducial and extract its position and angle

The most successful aplication that uses reacTIVision is the reactable, a novel musical instrument developed at the Music Technology Group of the Universitat Pompeu Fabra by the same team that develops reacTIVision.

The reactable is tabletop musical instrument that uses reacTIVision

reacTIVision is able to recognize fingers and objects tagged with a fiducial (something similar to a barcode). The purpose of my Master's Thesis, was to extract information from generic untagged objects.

Shape descriptors

The three selected shape descriptors are:

  • Elliptical approximation
  • Polygonal simplification
  • Medial Axis Transform (skeleton) simplification

Original binarized image

Elliptical approximation of the shape. It provides several control parameters: position, angle, area, ratio of principal axes

A polygonal approximation is useful for simple shapes recognition and visual feedback

The skeleton simplification is better suited for deformable objects

The skeleton is computed from the Distance Transform and the Voronoi Diagram

From the Distance Transform and the Voronoi Diagram we obtain the skeleton

The Distance Transform computes the distance to the nearest contour pixel. If we interpret the DT as a height in a 3D image, the ridges would be the skeleton:

The Distance Transform can be seen as this 3D surface where the height is the distance to the edge. The ridges in this surface are the Medial Axis Transform

This skeleton is then simplified keeping the points of high curvature and the necks.

Our skeleton simplification with several shapes. The key points are either in high curvature zones and necks of the shapes

From this simplification the shape can be reconstructed easily.

The original shape can be reconstructed from the skeleton simplification

Further information

You can watch my presentation or you can download the Master's Thesis from the MTG's web site or my local copy


Enter your comment
If you can't read the letters on the image, download this .wav file to get them read to you.
Back to top
shape_descriptors_for_tabletops.txt · Last modified: 2021/08/05 14:14 (external edit)
CC Attribution-Noncommercial-Share Alike 3.0 Unported = chi`s home Valid CSS Driven by DokuWiki Recent changes RSS feed Valid XHTML 1.0