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

GET-Forschungsseminar Abstracts

Perceptual Grouping Considering Depth Information for Artificial Visual Attention Based on Growing Neural Gas

Srikrishna Bashyam, GET Lab

25.11.2015, 16:30, P 1.4.17

Abstract:

In artificial vision, it is a challenge to extract and group elements in an image to form a perceptually meaningful object. In many applications, it is computationally more expensive to process the whole image compared to processing groups of pixels belonging to perceptually meaningful objects of interest. There are models of visual attention that select conspicuous locations in the scene. However, many models compute salient regions in the scene which often are only small parts of the target object. When identifying or tracking objects in technical systems such as autonomous robots, it is necessary to get the full object information instead of a single region. In GET Lab, a novel artificial visual attention system based on the Growing Neural Gas (GNG) was recently developed. In this approach, the resulting focus of attention (FOA) is a graph-based structure which is typically a part of an object but not the full object. This master thesis aims at developing a perceptual grouping for scene representations obtained by the Growing Neural Gas method. The focus of this work is to group these graph structures based on depth information which is obtained by stereo correspondences or by kinect cameras.