Hlavní stránka Info portálu ZČU

"Thanks to the newly developed algorithms, we will be able to better predict the movement of surrounding objects, which will contribute, for example, to increasing the safety and reliability of operation," said Jan Krejčí, a PhD student at the Department of Cybernetics at the Faculty of Applied Sciences of the University of West Bohemia in Pilsen (KKY FAV UWB), who was complemented by his colleague Jakub Matoušek: "The developed algorithms will be based on tensor decomposition (a tensor can be imagined as a multidimensional table), which will enable very efficient numerical solution of complex mathematical problems."

"The new algorithms will be so smart and trained that when tracking people on the street, they can work with a certain level of probability that the person being tracked may disappear from the camera's field of view for a certain period - for example, when they go to a shop, find themselves in a crowd of more people, or are momentarily obscured by a passing car. The algorithms will know that this person may reappear after some time and continue their route, so their trajectory needs to be tracked even when this person is not in the frame. When he reappears, they can recognize him and continue to follow his direction of travel," explained Jakub Matoušek, a doctoral student at KKY.

Current location-tracking algorithms are based on technologies from the 1970s. Although they are computationally inexpensive, they do not allow the full exploitation of the potential of modern sensors. The Pilsen and South Korean experts aim to develop modern methods that will provide more accurate data on tracked objects but keep computational and operational costs low.

As a result, they will offer an innovative solution for multi-object tracking with applications not only in autonomous mobility but also in medicine and biology - for example, for tracking the movement of cells or instruments in the hands of operating doctors, where the camera captures the movements of their hands. Such video footage helped, for example, medical students during the period of COVID-19 who could not attend operating theatres in person. At least they could experience the teaching of surgical techniques in this way.

The international team consists of Jindřich Duník, Jakub Matoušek, and Jan Krejčí from KKY and Marek Brandner from the Department of Mathematics (KMA) of FAV UWB, as well as Yeongkwon Choe from Kangwon National University and Chan Gook Park from Seoul National University.

The project 25-16919J Advanced State Estimation Methods in the Multi-target Tracking Task was supported by the GACR and the National Research Foundation of Korea (NRFK).