The Potential of Height Texture Measures for the Segmentation of Airborne Laser Scanner Data

Airborne laserscanning is being used for an increasing number of mapping and GIS data ac-quisition tasks. Besides the original purpose of digital terrain model generation, new applications arise in the automatic detection and modeling of objects such as buildings or vegetation for the generation of 3-D city models. A crucial prerequisite for the automatic extraction of objects on the Earth's surface from las-erscanner height data is the segmentation of datasets. Besides the height itself, height texture defined by local variations of the height is a significant feature of objects to be rec-ognized. The paper shows the potential of the analysis of height texture for the automatic segmentation of dense laserscanner datasets and the detection of objects in the segmented data. Based on the definition and computation of a number of texture measures used as bands in a classification approach, objects like buildings, single trees, ground vegetation and roads can be recognized. The technique was applied to a FLI-MAP laserscanner dataset with an average point density of more than five points per squaremeter, acquired over a village in The Netherlands. In a maximum likelihood classification based on the absolute height and several texture measures derived from it, a classification accuracy of 98 % could be achieved. Based on the result of the segmentation, all buildings of the scene and most single trees could be automatically detected. 1