Talk to be held at the workshop on image analysis and spatial statistics in forestry on November 2, 1999 at KVL, Frederiksberg, Denmark
Morten Larsen
(ml@dina.kvl.dk)
Dept. of Mathematics and Physics, Royal Veterinary and Agricultural University,
Frederiksberg, Denmark.
Templates derived from an optical tree crown model and from the geometry at image acquisition have been used with good results in the estimation of individual spruce tree top positions in aerial photographs. The method can find tree tops from a wide range of viewing angles including extreme off-nadir views, but two requirements have been that (1) the number of trees in a stand are known in advance and (2) the "average" 3D shape of the tree crowns is known.
A new development in the template matching technique is presented in which multiple templates arising from variation of the match window parameters "cast votes" to determine the most likely positions of tree tops. By analyzing the shape of the "voting curve" consisting of the tree top candidate vote counts sorted in decreasing order it is possible to (1) estimate the number of valid tree top candidates in the matching and (2) get an idea of how well the model tree crown shape corresponds to the actual tree crowns. The voting curve therefore eliminates the need for a separate method to estimate the number of trees in a stand and also allows an intelligent choice of the 3D crown shape parameters.