Your browser does not support JavaScript!

Home    Μια Εναλλακτική Προσέγγιση για τη Μελέτη της Αντίληψης του Χώρου: Χρήση Διατακτικού αντί Μετρικού Βάθους  

Results - Details

Add to Basket
[Add to Basket]
Identifier uch.csd.msc//1995lourakis
Title Μια Εναλλακτική Προσέγγιση για τη Μελέτη της Αντίληψης του Χώρου: Χρήση Διατακτικού αντί Μετρικού Βάθους
Alternative Title An alternative approach for studying space perception: Use of ordinal instead of metric depth
Author Λουράκης, Μανόλης Ι.Α
Thesis advisor Ορφανουδάκης, Στέλιος, 1948-2005
Abstract One of the most important capabilities of an intelligent system is that of navigation, that is the capability of autonomous motion in the environment, based on the measurements provided by various sensors. In order for such a system to achieve its goals, it should interact with dynamically changing and probably unknown surroundings. Consequently, the system must possess some sort of space perception, in other words possess capabilities that support various navigation functions. Undoubtedly, an important component of space perception is the perception of depth, the third world dimension that is lost during the projection of world parts on planar surfaces. Up to now, various methods have been used to compute the depth of points imaged in one or more frames. Information derived from more than one cameras, the motion of a camera with respect to the environment, the intensity of light reflected by the objects in an image and the distribution of some brightness patterns in an image, has formed the basis for the development of methods known with the generic name structure from X, where X can be one of stereo, motion, shading, texture. Traditionally, these methods aim at computing metric (quantitative) depth, in other words they aim at accurately estimating the distance of every point from the observer. The enormous and long-term effort towards computing metric depth, has made clear that this computation is very difficult. Although systems capable of computing metric depth in restricted cases have appeared, their generalization in general, real-world environments has not been successful. It is possible, however, that the understanding of some aspects of space does not necessarily require the computation of metric depth. Results of psychophysical experiments performed using human or animal subjects, suggest that an organism does not perceive the characteristics of the environment in extrinsic units (such as meters, seconds, etc) but employs units related to its body, motion capabilities and position in the environment instead. Moreover, the extracted information is qualitative rather than metric. Relying on such observations, this thesis proposes a qualitative approach for studying space perception and deals with the computation of ordinal depth. The term ``ordinal depth`` denotes the ordering of the points of a scene according to their depth, that is their distance from the observer. Clearly, ordinal depth provides less information compared to that provided by metric depth, nevertheless we believe that it can be very useful in understanding space. The approach that will be presented, aims at computing qualitative information, appropriate for space understanding, and follows the guidelines of the theory of purposive and qualitative active vision. In short, the method we developed for ordinal depth computation, uses a fixating image pair as input. Each image pair, acquired by the two cameras of a stereo system at a given time instance, can be seen as resulting from the movement of the one camera to the position of the other. This hypothetical motion and the approximation of the image formation process with the orthographic projection model, leads to simple motion equations for the image points, which permit the computation of the relationship of two points with respect to depth. A possible use of the resulting representation regards the qualitative description of the shape of surfaces contained in a scene.
Language Greek
Issue date 1995-04-01
Collection   School/Department--School of Sciences and Engineering--Department of Computer Science--Post-graduate theses
  Type of Work--Post-graduate theses
Views 384

Digital Documents
No preview available

Download document
View document
Views : 8