Division of Informatics
Forrest Hill & 80 South Bridge

Research Paper #449
Title:	Experiments with Segment-Based Stereo Using Dynamic Programming
Authors:	Trucco,E; Groppello,P; Burbello,F
Date:	Oct 1989
Presented:	Submitted to 1st Conference of the Associazione Italiana per l'Intelligenza Artificiale, 8-10 November, 1989, Trento
Keywords:
Abstract:	This paper addresses the problem of depth recovery using stereo analysis. The system described uses segments as features to be matched. A segment is defined as a group of connected edge points in the image, detected using the Marr-Hildreth technique. The identification of correspondent segments in the two images is obtained by a dynamic-programming based algorithm. After the detection phase each edge is described symbolically in terms of a list of attributes. The symbolic descriptions of the stereo pair are used for computing a correspondence cost for each candidate pair analysed. Only a subset of all possible matches in the feature space are considered. The set of matched pairs making the solution is given by the minimal cost path in the feature space. The output of the program is a disparity map of the scene. The system performed satisfactorily on a set of laboratory scenes. Experimental results are presented and discussed.
Download:	NO ONLINE COPY

Division of InformaticsForrest Hill & 80 South Bridge