Automatic Extraction of Trees from Aerial Imagery

We present two approaches for the automatic extraction of trees from high resolution aerial imagery. In the first approach we aim at isolated trees in rural areas. This approach is based on automatically generated digital surface models (DSM). It models trees as generalized ellipsoids of revolution which are fitted to the DSM. The parameters of the generated generalized ellipsoids of revolution are then used to determine if they represent deciduous or coniferous trees. In the second approach we try to extract leafless deciduous trees from images of urban areas captured in spring. We make use of the dark shadows of trees as well as of the fact that the vertical trunks are imaged as nadir pointing straight lines. The Hough transform is employed to generate hypotheses for the trunks. Branches are tracked by means of hysteresis thresholding. With this approach it is possible to determine the trunk base, height, width, and outline of the tree. First results show the feasibility of both approaches.