Model-based multi-sensor fusion

Abstract

A new class of ATR algorithms known in general as model-based have recently demonstrated some promising results. One major advantage of these new methods is that they are not only data-driven but they are also model-driven. The fact that they are model-driven removes the dependence of the recognition process on a low level segmentation process. These new model-based approaches accomplish segmentation and recognition simultaneously. In other words, the process entertains multiple segmentation possibilities until a sufficient amount of high level and low level information is gathered so that an intelligent interpretation of what the image contains can be formed. A second major advantage of these new methodologies is the fact that the fusion of information from multiple sensors is a continual process not a one shot deal which is typical of first generation algorithms. This paper will concentrate on the model-based ATR algorithm called relational template matching.