Model matching for SAR ATR based on probabilistic distance transforms

摘要

Abstract: In this paper we present an analysis of distance transform methods of matching object models to SAR data. We show that by properly defining the distance function, the likelihood of each observed SAR feature data point given the model is given as a function of position. This allows calculation of a likelihood of observing a set of data features, given a model and its associated pose and other parameters. The issue of normalization resulting from the non-correspondence based distance transform method is discussed. When prior densities of the model features are available, maximum a- posteriori results are obtainable. This method allows the use of priors of models and individual features, along with the geometric probability densities associated with the feature prediction and measurement processes, to be incorporated within a fast correlation-type distance transform matching module. The method also potentially allows exploitation of persistent scatterers over a limited range of SAR model-to-target imaging parameters.!12