The Detection and Identification of Man-Made Objects from Aerial Reconnaissance Photographs

摘要

An algorithm is proposed which detects and identifies man-made objects from actual aerial reconnaissance photographs. The objects are detected by cross-correlating the contour map of the digitized photograph against a generalized model for all man-made objects. This model, consisting of only three prototypes, detects objects regardless of their size, shape, and orientation. The boundary of each man-made object is then traced and represented as a parametric set of complex points. This representation is then cross-correlated with prototypes representing different general shapes in order to gain information about the size and shape of the object. This information is then combined with a logical decision process, based on some generalized rules, to classify the objects. The results of this investigation are based on a test set of six actual photographs. The procedure exhibited a detection accuracy of 71% and an identification accuracy of 87%. (Author)