Domestic service robots are becoming increasingly available and affordable. However, interaction with the real world is still a challenge. Tracker systems are fast and can yield good results because they exploit temporal coherence, but often need manual initialization. In this research such a tracker system was combined with a detector system that initializes and corrects the tracker automatically. The system was used to track faces in a dataset. Different (combinations of) features were explored. The combined system was also used to guide a mobile, autonomous robot. Data was gathered as the robot performed its task. This data was used to train the tracker system, to make it more robust to changes in the environment, like changing illumination conditions. This robustness is necessary for the tracking system to handle the changing conditions in the real world without a lot of feedback from an external system. The combined detection and tracking system was compared to the detector system is isolation. The combined system was able to run faster or perform better, depending on its settings. Adaptive combination rules helped to choose the best feature(s) for a certain situation. In an evaluation on data gathered in the real world, the trained tracker performed better than the non-learning tracker.
This photo report gives you a look behind the scenes of the work at the Ocean Grazer project.
The festive opening of the Bernoulli Institute for Mathematics, Computer Science and Artificial Intelligence (UG) will be held on 1 November, with a Symposium that will combine pitches of interdisciplinary research at the Bernoulli, poster sessions...
Gosens wins the Prix Galien Research Award 2018