Abstract

Automatic pattern recognition using multiple sensors offers a promising approach to many pattern recognition problems. Among the many research issues which remain are determination and extraction of useful information from the imagery and fusion of the information derived from multiple sources. The technique being investigated applies the Dempster-Shafer theory of evidence¹ to information in the form of features extracted from multiple sensor views of a scene. Features are extracted from segmented regions of a database of real corresonding IR and absolute range images. Multiple sensor features are represented through mass functions for fusion using Dempster’s rule of combination. The resulting mass distributions are evaluated to determine an estimate of class membership for each region.

© 1988 Optical Society of America