%0 Generic
%A Höfle, Bernhard
%C Innsbruck
%D 2007
%F heidok:36930
%R 10.11588/heidok.00036930
%T Detection and utilization of the information potential of airborne laser scanning point cloud and intensity data by developing a management and analysis system
%U https://archiv.ub.uni-heidelberg.de/volltextserver/36930/
%X In recent years Airborne Laser Scanning (ALS) evolved to the state-of-the-art technique for topographic data acquisition. High resolution and highly accurate digital elevation models can be produced with a high degree of automation. Especially in vegetated areas and in areas with low surface texture this active remote sensing technique surpasses traditional aerial imagery in acquiring terrain information.    Until now, primarily the rasterized elevation models (Digital Terrain and Surface Models) were used due to the simple data model (regular grid) and the multitude of existing image/raster analysis methods. The original measurements – the ALS point cloud – are generally delivered to end-users but not often used because of the lack of management and analysis strategies that can handle the incidental data volume (>1 pt/m2 multiplied by scanned area). The point cloud – tuples of 3D coordinates, a value for received signal intensity and a timestamp – contains the highest degree of information as it represents the three-dimensional, ungeneralized result of ALS data acquisition.    This thesis aims to utilize the full information inherent in the ALS point cloud and the intensity value. Therefore, a new concept for spatial data management of laser points and their additional information (e.g. intensity, timestamp, scanner positions) is introduced. The Geographic Information System (GIS) is used to bridge existing raster processing methods and novel point cloud management and analysis methods. The developed information system “LISA” (LIdar Surface Analysis) integrates existing Open Source software (e.g. GRASS GIS) allowing a fast and easy implementation of new processing tools, which can make use of both, ALS elevation models and the original point cloud. Point cloud management is performed in an external spatial database management system (PostgreSQL/PostGIS). A new ALS data model was developed and the existing spatial index was extended to improve performance.    Previous research studies working with signal intensity of pulsed ALS systems stated that there is a large potential for supporting surface classification and object detection. Before making use of the intensity value, it has to be corrected for known influences (e.g. emitted power, spherical loss, topographic and atmospheric effects), which are derived from the radar equation. This thesis introduces, implements and evaluates two fully automatic approaches for intensity correction to derive a value proportional to surface reflectance, which is well suited for surface classification. The first method, the data-driven approach, empir-  ically determines a correction function accounting for influences correlating with range (distance scanner-target). The second method, the model-driven approach, derives individual correction factors for each laser point by applying the radar equation. The “disturbing” effects could be successfully reduced (<noise level of ca. 10%) with both methods.    Two application examples demonstrate the large potential of having concurrent access to the point cloud and its derivatives (e.g. Digital Surface Model, corrected intensity, laser range vectors). The GIS component of the LISA system allows implementing object-based analysis approaches, which overcome the “salt & pepper” problem of traditional image analysis. Additionally, geometric, topological, contextual and multi-resolution information can be regarded in the object-based methodology.    The first application example detects glacier features on basis of the original ALS point cloud. A segmentation task (region growing) utilizes the spatial and corrected intensity information of the laser points, which exhibits a good separability of the different glacier surface facies (ice, firn, snow and surface irregularities). Two-dimensional alpha shapes and a gap filling procedure (using Voronoi diagrams) are used to derive a gapless vector representation (GIS vector layer) of the glacier surface. A supervised, rule-based classification is applied to label the homogeneous segments. Further, neighboring segments of the same class are merged to the final objects. An overall classification accuracy of >90% and a delineation accuracy <2 m could be achieved.    The second application introduces a novel procedure for building detection and roof facet modeling and classification by combining object-based image and point cloud analysis. This thesis contributes the framework (LISA system) for point cloud analysis put on top of the building detection algorithm. The final roof modeling originates from a hierarchical, multi-resolution object detection. Firstly, the building polygons are derived from a fill sinks segmentation on the inverted Digital Surface Model. Secondly, these building objects are used to extract the corresponding laser points for each building. Finally, a point cloud segmentation task (region growing) using the normal vectors in each laser point (derived by plane fitting in a defined neighborhood) is applied to identify planar objects (roof patches). The 3D roof facets are further classified by their slope. This result can be used as input for a multitude of applications, such as snow load capacity modeling or photovoltaics site analysis.    This thesis concludes that efficient ALS data management integrated in a GIS environment allows moving some steps forward to the major goal of the “detection and utilization of the the full information potential”. Without additional costs added value can be generated out of existing ALS datasets (e.g. intensity correction). Future research should incorporate state-of-the-art ALS data (full-waveform) as well as terrestrial laser scanning data.