• Home
  
  
• Übersicht
• Forschung
• Lehre
• Mitarbeitende
• Publikationen
• Auszeichnungen
• Sonstiges
  
  
• Aktuelles
• Kontakt

GET-Forschungsseminar Abstracts

Scene Gist Extraction for Visual Attention Based on Growing Neural Gas

Rohan Sukumaran, GET Lab

13.04.2016, 16:30, P 1.4.17

Abstract:

Autonomous mobile robots operating in outdoor environments rely heavily on visual data for efficient navigation and interaction. Processing and retrieving meaningful and accurate information from visual data can be time-consuming and requires a lot of computational resources. For efficiently processing visual data, artificial visual attention systems are employed. Artificial visual attention systems concentrate on guiding attention to important regions of the image. For the identification of the most important regions in an image, artificial visual attention systems often compute salient regions. In GET Lab, a novel artificial visual attention system is based on the Growing Neural Gas (GNG), which uses self-organizing graph structures to approximate regions in the image for computing saliency. To achieve more accuracy and efficiency, knowledge about the environment can help to prune unimportant details at a very early stage. Having knowledge about the type of the scene is vital to extracting its gist. This thesis concentrates on a novel approach using the GNG to obtain a rough representation for gist and refining it for a biased saliency computation and guidance visual attention.