Extraction of tree groups from high-resolution digital surface models

Abstract

Digital surface models (DSMs) represent a complementary data source to images based on spectral reflectances in remote sensing information processing. Especially high resolution DSMs (sub-meter scale) offer a level of detail sufficient for most forest interpretation tasks. One major drawback in DSM data is the inherent union of object and terrain height. This aggravates image processing tasks like object extraction. It is the aim of this paper to overcome this drawback and solve the problem of extracting borders of tree groups as well as their heights above the ground on the basis of a DSMs of forested areas. The presented approach can be subdivided as follows: first the original DSM is normalized to a preliminary above-ground-model by comparing all pixels to a minimum value in their neighborhood. Non-ground points are identified by thresholding this model. The ground elevation of the non-ground regions is interpolated based on the elevation of the pixels neighboring these regions. Subtracting the resulting DTM from the original DSM yields precise object heights. Finally, a contouring algorithm is applied to retrieve the borders of the tree groups.