Optimizing templates for finding trees in aerial photographs

Positions of individual trees in high-resolution aerial photographs are estimated by use of templates derived from an optical tree crown model and the geometry at image acquisition. The template is bounded by an ellipse, for which we find the optimal size, eccentricity and translation along one axis. Tree top positions are estimated as local maxima of the correlation function for the successively transposed template over the whole image. Performance is evaluated using three images from a thinning experiment in spruce. The templates are compared with empirical shapes obtained by averaging over the individual trees in the image. (C) 1998 Elsevier Science B.V. All rights reserved.