Cooperative integration of multiple stereo algorithms

Abstract
A novel scheme is proposed to integrate multiple stereo algorithms in a cooperative framework. Each algorithm is implemented in a separate module and executed in parallel. The stereo correspondence obtained in each algorithm is stored in the module. Confidence of each stereo correspondence is evaluated based on the ambiguity in the search process and it is attached to the correspondence result. During the execution, each module communicates with other modules and offers subsets of its results to another by its request. The cooperation among the modules makes it possible not only to improve the performance of each algorithm but also to make the integrated system highly adaptive to various scenes. A system integrating three stereo matching algorithms which use different kinds of image features has been developed on a parallel computer to demonstrate the feasibility of the scheme.

This publication has 7 references indexed in Scilit: