Hlavní stránka Info portálu ZČU

Let's imagine that a doctor in surgery will be able to record our movements in detail, and algorithms will then evaluate whether we are performing the movement correctly or whether it shows any pathological phenomena. Or we could use virtual reality to look under the hands of assemblers and create video instructions on how to assemble, for example, a cabinet correctly. The results of the research being worked on by scientists from the Faculty of Applied Sciences at the University of West Bohemia in Pilsen (FAV UWB), which aims to improve computer recognition of movements, will have wide application in areas where accurate modeling of moving objects plays a key role.

More effective use of dynamic 3D models will thus be possible in academia and industry, as well as in developing PC games and animated films. In addition to improving computer recognition of 3D object motion, the improved algorithms will also reduce the data used. The team led by Libor Váša from the Department of Computer Science and Computer Engineering at FAV UWB has been working on that and received support for the research from the Grant Agency of the Czech Republic.

And what exactly does better and more economical motion recording consist of? Modern technologies such as cameras, cameras, or depth sensors allow the faithful capture of accurate three-dimensional shapes in motion - for example, human movement, fabric deformation, or facial expressions. Existing algorithms typically represent such an object as a sequence of point clouds in space that form a 3D shape together. "But unlike a human, a computer cannot distinguish which parts in one frame correspond to parts in the next frame. It doesn't know that a given point cloud represents the same part of an object, even though those points might be located elsewhere. The project aims to teach computers to do just that to recognize objects in all scenes - faces, hands, feet - that are moving and deforming somehow. This data representation is much more economical and suitable for further analysis," explained Libor Váša.

The project was developed in collaboration with the Università della Svizzera italiana in Lugano. Professor Kai Hormann, one of the world's leading experts on barycentric coordinates and cage deformations, the mathematical concepts on which the tracking of deforming spatial objects is based, will also be involved in developing the new solutions. The combination of his expertise with the Czech team will enable the creation of new algorithms for the representation of dynamic shapes and the optimization of their computational processing.

The international project "Coordinate Representation of Time-Varying Triangular Networks" was financially supported by GAČR and the Swiss National Science Foundation. Project number: 25-16495K.