Visual attention and active vision. From natural to artificial systems

Promotie: dhr. G.W. Kootstra, 13.15 uur, Academiegebouw, Broerstraat 5, Groningen

Proefschrift: Visual attention and active vision. From natural to artificial systems

Promotor(s): prof.dr. L.R.B. Schomaker

Faculteit: Wiskunde en Natuurwetenschappen

Contact: Gert Kootstra, tel. +46 08790 6792

Visual attention and active vision. From natural to artificial systems

This dissertation deals with visual attention and active vision in man and machine. The topics are studied in human perception, which gives inspiration to develop robotic systems. Vision, instead of being passive, involves active processes to focus attention. This is not only true for natural systems, but it is also important for artificial systems. This dissertation proposes symmetry, one of the Gestalt principles for figure-ground segregation, as an important feature to guide visual attention.

In the first part of the dissertation, the perception of symmetry by the human visual system is studied in the context of overt visual attention. We propose a visual attention model based on symmetry. The results show that human eye fixations are predicted better by our model than by a model using center-surround contrasts of basic features such as brightness, color, and orientation. The results furthermore indicate that symmetry is detected efficiently by humans, despite being a higher-level visual feature.

In the second part, the proposed visual attention model is applied to focus the attention of a robot on interesting parts in its environment. The use of symmetry is shown to be beneficial for the selection of stable and robust visual landmarks. Based on these landmarks, the robot builds a map of the environment and uses this map to localize itself. In this context, the use of symmetry outperforms the landmark selection method based on center-surround contrasts. It is furthermore illustrated that perception is simplified using active vision.