Title:	Self-Organising Maps for Efficient Data Reduction and Visual Optimisation of Stereoscopic based Disparity Maps
Authors:	Müller, Simone Kranzlmüller, Dieter
Citation:	WSCG 2022: full papers proceedings: 30. International Conference in Central Europe on Computer Graphics, Visualization and Computer Vision, p. 227-234.
Issue Date:	2022
Publisher:	Václav Skala - UNION Agency
Document type:	conferenceObject
URI:	http://hdl.handle.net/11025/49598
ISBN:	978-80-86943-33-6
ISSN:	2464-4617
Keywords:	Kohonenova síť;samoorganizující síť;hloubková segmentace obrazu;disparitní mapy;počítačové vidění
Keywords in different language:	Kohonen networks;self-organizing map;depth image segmentation;disparity maps;computer vision
Abstract in different language:	Many modern autonomous systems use disparity maps for recognition and interpretation of their environment. The depth information of these disparity maps can be utilised for point cloud generation. Real-time and high-quality processing of point clouds is necessary for reliable detection of safety-relevant issues such as barriers or obstacles in road traffic. However, quality characteristics of point clouds are influenced by properties of depth sensors and environmental conditions such as illumination, surface and texture. Quality optimisation and real-time implemen- tation can be resource intensive. Limiting the amount of data allows optimisation of real-time processing. We use Kohonen network existing self-organising maps to identify and segment salient objects in disparity maps. Kohonen networks use unsupervised learning to generate disparity maps abstracted by a small number of vectors instead of all pixels. The combination of object-specific segmentation and reduced pixel number decreases the memory and processing time towards real-time compatibility. Our results show that trained self-organising maps can be applied to disparity maps for improved runtime, reduced data volume and further processing of 3D reconstruction of salient objects.
Rights:	© Václav Skala - UNION Agency
Appears in Collections:	WSCG 2022: Full Papers Proceedings

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.