This paper highlights a novel segmentation approach for single treesfrom LIDAR data and compares the results acquired both from first/lastpulse and full waveform data. In a first step, a conventional watershed-basedsegmentation procedure is set up, which robustly interpolates thecanopy height model from the LIDAR data and identifies possible stempositions of the tallest trees in the segments calculated from thelocal maxima of the canopy height model. Secondly, this segmentationapproach is combined with a special stem detection method. Stem positionsin the segments of the watershed segmentation are detected by hierarchicallyclustering points below the crown base height and reconstructingthe stems with a robust RANSAC-based estimation of the stem points.Finally, a new three-dimensional (3D) segmentation of single treesis implemented using normalized cut segmentation. This tackles theproblem of segmenting small trees below the canopy height model.The key idea is to subdivide the tree area in a voxel space and toset up a bipartite graph which is formed by the voxels and similaritymeasures between the voxels. Normalized cut segmentation dividesthe graph hierarchically into segments which have a minimum similaritywith each other and whose members (= voxels) have a maximum similarity.The solution is found by solving a corresponding generalized eigenvalueproblem and an appropriate binarization of the solution vector. Experimentswere conducted in the Bavarian Forest National Park with conventionalfirst/last pulse data and full waveform LIDAR data. The first/lastpulse data were collected in a flight with the Falcon II system fromTopoSys in a leaf-on situation at a point density of 10 points/m2.Full waveform data were captured with the Riegl LMS-Q560 scannerat a point density of 25 points/m2 (leaf-off and leaf-on) and ata point density of 10 points/m2 (leaf-on). The study results provethat the new 3D segmentation approach is capable of detecting smalltrees in the lower forest layer. So far, this has been practicallyimpossible if tree segmentation techniques based on the canopy heightmodel were applied to LIDAR data. Compared to a standard watershedsegmentation procedure, the combination of the stem detection methodand normalized cut segmentation leads to the best segmentation resultsand is superior in the best case by 12%. Moreover, the experimentsshow clearly that using full waveform data is superior to using first/lastpulse data.     «

This paper highlights a novel segmentation approach for single treesfrom LIDAR data and compares the results acquired both from first/lastpulse and full waveform data. In a first step, a conventional watershed-basedsegmentation procedure is set up, which robustly interpolates thecanopy height model from the LIDAR data and identifies possible stempositions of the tallest trees in the segments calculated from thelocal maxima of the canopy height model. Secondly, this segmentationapproach is combin...     »