Automatic estimation of individual tree positions from aerial photos

An automatic method is suggested for estimating the positions of individual trees from panchromatic aerial photos of even-aged homogeneous stands of Norway spruce (Picea abies (L.) Karst.). The scanned photo is smoothed by a two-dimensional isotropic Gaussian kernel such that the number of local maxima above the most frequent grey level is approximately equal to the number of trees. The positions of trees at ground level are estimated by use of a displacement model incorporating the angle to the sun, the camera position, and estimated tree heights. Parameters of the model are estimated from data for a thinning experiment with six thinning treatments and field measurements of the individual tree positions at ground level. The displacement model is used for matching trees and maxima. It is found that for medium and heavy thinning about 95%, and for light thinning about 85%, of the trees can be detected and that the root mean square residual error in the displacement model is about 65 cm. For estimation of tree positions at ground level additional errors due to uncertainties in the heights may become important, in particular for trees far from the nadir point. Resume : On propose une mOthode automatisOe permettant diestimer la position diarbres individuels ‡ partir de photos aOriennes panchromatiques de peuplements Oquiennes diOpinettes de Norvge ( Picea abies (L.) Karst.). La photo balayOe est lissOe au moyen diun algorithme gaussien isotropique bi-dimensionnel qui fait que le nombre des maxima locaux supOrieurs au ton de gris le plus frOquent Oquivaut approximativement au nombre diarbres. La position des arbres au sol est estimOe en