Search

In this paper, different methods for the evaluation of building detection algorithms are compared. Whereas pixel-based evaluation gives estimates of the area that is correctly classified, the results are distorted by errors at the building outlines. These distortions are potentially in an order of 30%. Object-based evaluation techniques are less affected by such errors. However, the performance metrics thus delivered are sometimes considered to be less objective, because the definition of a ldquocorrect detectionrdquo is not unique. Based on a critical review of existing performance metrics, selected methods for the evaluation of building detection results are presented. These methods are used to evaluate the results of two different building detection algorithms in two test sites. A comparison of the evaluation techniques shows that they highlight different properties of the building detection results. As a consequence, a comprehensive evaluation strategy involving quality metrics derived by different methods is proposed.

Since 2003, the full-waveform laser altimetry system onboard NASA's Ice, Cloud and land Elevation Satellite (ICESat) has acquired a worldwide elevation database. ICESat data are widely applied for change detection of ice sheet mass balance, forest structure estimation, and digital terrain model generation of remote areas. ICESat's measurements will be continued by a follow-up mission. To fully assess the application possibilities of the full-waveform products of these missions, this research analyzes the vertical accuracy of ICESat products over complex terrain with respect to land cover type. For remote areas, validation of individual laser shots is often beyond reach. For a country with extensive geo-infrastructure such as The Netherlands, excellent countrywide validation is possible. Therefore, the ICESat full-waveform product GLA01 and the land elevation product GLA14 are compared to data from the Dutch airborne laser altimetry archive Actual Height model of the Netherlands (AHN). For a total population of 3172 waveforms, differences between ICESat- and AHN-derived terrain heights are determined. The average differences are below 25 cm over bare land and urban areas. Over forests, differences are even smaller but with slightly larger standard deviations of about 60 cm. Moreover, a waveform-based feature height comparison resulted in feature height differences of 1.89 m over forest, 1.48 m over urban areas, and 29 cm over low vegetation. These results, in combination with the presented processing chain and individual waveform examples, show that state-of-the-art ICESat waveform processing is able to analyze waveforms at the individual shot level, particularly outside urban areas.

Within the thematic area of Databases and Modelling a certain focus is placed on the effective acquisition and management of geo-data and the derivation of standardized products from this data, e.g. as input parameters in process simulation models. In most cases, in-situ data collection (e.g. run-off measurements) is state-of-the-art. As there is a growing demand for area-wide data collection, the utilization of remote sensing technology will gain ground in the future. Within alpS the project ‘Determination of surface properties from laser scanning data’ addresses these demands by incorporating certain aspects of remote sensing in natural hazard management. Remote sensing is the contactless collection of information about an object or process. This is done with electromagnetic waves and imaging methods. Remote sensing can be carried out from the earth’s surface or from airborne and spaceborne platforms. For earth observation issues passive sensors are widely applied, which record the reflected radiation of natural energy sources (with the sun as the most important one). For many applications the method of choice is still aerial photography. Active sensors are more flexible as they have their own energy source. Laser scanning is such an active method.

Airborne LiDAR (Light Detection And Ranging) combines cost efficiency, high degree of automation, high point density of typically 1–10 points per m2 and height accuracy of better than ±15 cm. For all these reasons LiDAR is particularly suitable for deriving precise Digital Terrain Models (DTM) as geometric basis for hydrodynamic-numerical (HN) simulations. The application of LiDAR for river flow modelling requires a series of preprocessing steps. Terrain points have to be filtered and merged with river bed data, e.g. from echo sounding. Then, a smooth Digital Terrain Model of the Watercourse (DTM-W) needs to be derived, preferably considering the random measurement error during surface interpolation. In a subsequent step, a hydraulic computation mesh has to be constructed. Hydraulic simulation software is often restricted to a limited number of nodes and elements, thus, data reduction and data conditioning of the high resolution LiDAR DTM-W becomes necessary. We will present a DTM thinning approach based on adaptive TIN refinement which allows a very effective compression of the point data (more than 95% in flood plains and up to 90% in steep areas) while preserving the most relevant topographic features (height tolerance ±20 cm). Traditional hydraulic mesh generators focus primarily on physical aspects of the computation grid like aspect ratio, expansion ratio and angle criterion. They often neglect the detailed shape of the topography as provided by LiDAR data. In contrast, our approach considers both the high geometric resolution of the LiDAR data and additional mesh quality parameters. It will be shown that the modelling results (flood extents, flow velocities, etc.) can vary remarkably by the availability of surface details. Thus, the inclusion of such geometric details in the hydraulic computation meshes will gain importance for river flow modelling in the future.

Three dimensional city models are necessary for supporting numerous management applications. For the determination of city models for visualization purposes, several standardized workflows do exist. They are either based on photogrammetry or on LiDAR or on a combination of both data acquisition techniques. However, the automated determination of reliable and highly accurate city models is still a challenging task, requiring a workflow comprising several processing steps. The most relevant are building detection, building outline generation, building modeling, and finally, building quality analysis. Commercial software tools for building modeling require, generally, a high degree of human interaction and most automated approaches described in literature stress the steps of such a workflow individually. In this article, we propose a comprehensive approach for automated determination of 3D city models from airborne acquired point cloud data. It is based on the assumption that individual buildings can be modeled properly by a composition of a set of planar faces. Hence, it is based on a reliable 3D segmentation algorithm, detecting planar faces in a point cloud. This segmentation is of crucial importance for the outline detection and for the modeling approach. We describe the theoretical background, the segmentation algorithm, the outline detection, and the modeling approach, and we present and discuss several actual projects.

In recent years object-based image analysis of digital elevation models acquired by airborne laser scanning gained in importance. Various applications for land cover classification (e.g. building and tree detection) already show promising results. Additionally to elevation rasters the original airborne laser scanning point cloud contains highly detailed 3D information. This paper introduces an integrative approach combining object-based image analysis and object-based point cloud analysis. This integrative concept is applied to building detection in the raster domain followed by a 3D roof facet delineation and classification in the point cloud. The building detection algorithm consists of a segmentation task, which is based on a fill sinks algorithm applied to the inverted digital surface model, and a rule-based classification task. The 340 buildings of the test site could be derived with 85% user’s accuracy and 92% producer’s accuracy. For each building object the original laser points are further investigated by a 3D segmentation (region growing) searching for planar roof patches. The finally delineated roof facets and their descriptive attributes (e.g. slope, 3D area) represent a useful input for a multitude of applications, such as positioning of solar-thermal panels and photovoltaics or snow load capacity modeling.

Analysis of full-waveform pulses from space-based laser altimeter systems are expected to improve our ability of measuring forests globally. Moreover, with the increase in the number of waveform data sets, it is now possible to study temporal changes in waveform returns over the same spatial domain. ICESat full waveform data from two epochs, i.e. winter and summer (2003) along near-coincident ground tracks, are studied. Data analysis methods are discussed, including normalization and matching of near-coincident waveforms, Gaussian decomposition, and derivation of forest measurement and forest change parameters. We quantify differences between winter and summer waveforms, acquired over broad-leaved, mixed-wood, and needle-leaved forests in Europe. The results indicate that, although maximum tree height barely changes over six months, i.e.

Airborne laser scanning (ALS) is a remote sensing technique well-suited for 3D vegetation mapping and structure characterization because the emitted laser pulses are able to penetrate small gaps in the vegetation canopy. The backscattered echoes from the foliage, woody vegetation, the terrain, and other objects are detected, leading to a cloud of points. Higher echo densities (> 20 echoes/m2) and additional classification variables from full-waveform (FWF) ALS data, namely echo amplitude, echo width and information on multiple echoes from one shot, offer new possibilities in classifying the ALS point cloud. Currently FWF sensor information is hardly used for classification purposes. This contribution presents an object-based point cloud analysis (OBPA) approach, combining segmentation and classification of the 3D FWF ALS points designed to detect tall vegetation in urban environments. The definition tall vegetation includes trees and shrubs, but excludes grassland and herbage. In the applied procedure FWF ALS echoes are segmented by a seeded region growing procedure. All echoes sorted descending by their surface roughness are used as seed points. Segments are grown based on echo width homogeneity. Next, segment statistics (mean, standard deviation, and coefficient of variation) are calculated by aggregating echo features such as amplitude and surface roughness. For classification a rule base is derived automatically from a training area using a statistical classification tree. To demonstrate our method we present data of three sites with around 500,000 echoes each. The accuracy of the classified vegetation segments is evaluated for two independent validation sites. In a point-wise error assessment, where the classification is compared with manually classified 3D points, completeness and correctness better than 90% are reached for the validation sites. In comparison to many other algorithms the proposed 3D point classification works on the original measurements directly, i.e. the acquired points. Gridding of the data is not necessary, a process which is inherently coupled to loss of data and precision. The 3D properties provide especially a good separability of buildings and terrain points respectively, if they are occluded by vegetation.

After a short introduction into the Airborne Laser Scanning (ALS), the sampling process and a brief comparison of ALS with other data acquisition methods, the paper focuses on the generation of high quality digital terrain models from ALS data. In the first part the hierarchic robust interpolation method for the automatic filtering and classification, respectively, of an ALS point cloud into terrain and off-terrain (points on buildings or on the vegetation, etc.) is presented. Then, an approach for the modelling of linear features, such as break lines, based on the ALS data is described. Based on the feature lines and the classified terrain points a high quality DTM can be determined. For many subsequent processes the data reduction of the detailed DTM is essential in order to handle the large amount of data. Therefore, the paper presents a data reduction method that guaranties a certain DTM representation quality. Next to the presented algorithms, for each of the mentioned topics, references to other published methods are provided. Furthermore, beside the description of the algorithms, practical examples demonstrate the results of the algorithms. At the end of the paper, an outlook section discusses the advanced possibilities of DTM determination based on full-waveform ALS data.

In this overview paper the principles of laser scanning systems are presented. This includes a survey of different range measurement principles as well as different mechanisms for the deflection of the emitted laser beam. Furthermore, the usage of the laser scanning (LS) principle at different platforms (airborne (ALS), terrestrial (TLS), satellite) is discussed. Additionally, typical sensor parameters of currently commercially available sensor systems are presented. Furthermore, the technique of full-waveform LS is introduced and georeferencing of ALS data for improved precision is presented. The usage of LS data in different applications will be presented in an overview.

Full waveform laser altimetry is a recently developed method to obtain a complete vertical profile of the height of objects in the footprint as illuminated by a laser pulse. The richness of the signal also complicates the processing. One way to improve the processing strategy is to analyze differences of waveforms that should be very similar because they were obtained at approximately the same time and location. Such waveform pairs are still difficult to find. Here it is shown how to use the archive of ICESat space-borne altimetry data over Europe to determine a set of tenths of thousands of at least partial overlapping waveform pairs. The differences in the values of the waveform parameters, median energy, waveform extent, relative returned energy and intensity distribution are determined and discussed. As a case study, three typical pairs of almost perfectly overlapping waveforms are shown, were considerable differences are still occurring. In all three cases an explanation for these differences is found and discussed. Further analysis of the waveform pairs in this database is expected to considerably improve automatic processing of full waveform data.

Multi beam echo sounding is the state of the art way for surveying sea bottoms. The sea floor elevation is obtained strip wise by measuring the time it takes for sound signals, emitted simultaneously in different directions, to travel to the sea bottom and back. We compare various ways of filtering erroneous soundings from MBES data sets, all based on Kriging. This research was initiated because of the problems a classic 1D cross validation method had with distinguishing blunders from features, like pipelines. We show that part of the problems can be solved by extending the 1D method to 2D and that most problems are solved by a robust, iterative filter method.

Landslides play a key role in landscape evolution in the Eastern Alps. These geomorphic phenomena are influenced by multiple interdependent and interacting natural and anthropogenic factors. An in-depth evaluation of the spatial distribution of existing landslides enables to gain first insights into potentially hazardous areas. Morphometric analysis techniques of mapped landslides as well as their date of occurrence allow to infer their activity and also potential impacts on affected areas. The prevalent slow moving landslides and inactive slipping areas were mapped and analysed via digital terrain models (DTM), shaded relief images of highly resolved airborne laserscanning (ALS) data and in-field observations. Orthophotos from aerial surveys and ALS data allowed a deferred-time analyses of past landslide occurrences including a record of recent slope movements. All mapped landslides were classified and analysed with geomorphometric indices. Pedological processes, the lithological setting and anthropogenic landscape transformationwere taken into account when interpreting the results. The geomorphometrical evaluation of the sliding areas determine the creation of a multitemporal landslide inventory in the Northern Walgau.

Analysis of common neighborhood definitions for airborne laser data, triangulation or raster-based, reveals deficiencies in modeling the measured objects. Concepts that originate from 2D data structures are used for modeling complex 3D objects and for handling datasets with different point densities. Realizing these shortcomings, this paper proposes a new neighborhood system for airborne laser data. Based on laser data characteristics the proposed systems consider, among other features, point density, layered and overhanging structures, and local surface trends. Parameters for the proposed systems are derived from theoretical and practical observations. The paper demonstrates the type of neighborhood that is established by the proposed systems, and shows that artifacts that are usually created by the common neighborhoods are avoided here, and that structures within the data that are usually masked are revealed. The paper demonstrates how subsequent applications benefit from the new system. Finally, the estimation of surface normals by the proposed systems is compared to the triangulation; results show a significant improvement in the reliability and quality of the estimation.

In this article we extend our previous work on the topic of ALS strip adjustment without GNSS/IMU trajectory data. Between overlap- ping strip pairs the relative orientation as a 3D affine transformation is estimated by a 3D LSM approach, which uses interpolated 2.5D grid surface models of the strips and the entire strip overlap as one big LSM window. The LSM derived relative orientations of all strip pairs in the block together with their covariance matrices are then used simultaneously as observations in an adjustment following the Gauss-Helmert model. This way the exterior orientation of each strip is computed, which refers to a relative block system. If proper ground control data is given, then an absolute orientation of the block of strips can be computed by a final LSM run. In a small example consisting of 4 strips with ca. 70% overlap the improvement in the relative geometric accuracy is demonstrated by the decreasing óMAD of the height differences from 8.4cm (before) to 1.6cm (after the strip adjustment).

The first processing step of Light Detection and Ranging (LiDAR) data is the orientation of the acquired point cloud. This applies to the terrestrial as well as to the airborne case. In the terrestrial application, registration, that is alignment to a common reference frame, must be distinguished from geo-referencing. A reference to a geodetic coordinate system is not always needed in terrestrial applications, as sometimes a local coordinate system is sufficient. Special attention is paid to registration approaches that require neither markers nor additional orientation sensors. In the airborne case, a simultaneous orientation and determination of calibration parameters on the job is becoming standard. In the airborne case, the classification into ground points (or pixels for elevation images) and off-terrain is also standard procedure and necessary for almost all applications. © 2008 Taylor & Francis Group.

In this contribution the complexity of the vertical vegetation structure, based on dense airborne laser scanning (ALS) point cloud data (25 echoes/m2 ), is analyzed to calculate vegetation roughness for hydraulic applications. Using the original 3D ALS point cloud, three levels of abstractions are derived (cells, voxels and connections) to analyze ALS data based on a 1×1 m2 raster over the whole data set. A voxel structure is used to count the echoes in predefined detrended height levels within each cell. In general, it is assumed that the number of voxels containing echoes is an indicator for elevated objects and consequently for increased roughness. Neighboring voxels containing at least one data point are merged together to connections. An additional height threshold is applied to connect vertical neighboring voxels with a certain distance in between. Thus, the connections indicate continuous vegetation structures. The height of the surface near or lowest connection is an indicator for hydrodynamic roughness coefficients. For cells, voxels and connections the laser echoes are counted within the structure and various statistical measures are calculated. Based on these derived statistical parameters a rule-based classification is developed by applying a decision tree to assess vegetation types. Roughness coefficient values such as Manning's n are estimated, which are used as input for 2D hydrodynamic-numerical modeling. The estimated Manning’s values from the ALS point cloud are compared with a traditional Manning's map. Finally, the effect of these two different Manning's n maps as input on the 2D hydraulics are quantified by calculating a height difference model of the inundated depth maps. The results show the large potential of using the entire vertical vegetation structure for hydraulic roughness estimation.

The influence of low vegetation on laser scanning and its disturbing effect (a systematic positive height shift) during DTM generation is well recognized. Based on our experience in a previous study of estimating the effect of medium-height vegetation (shrubs and bushes) in rough terrain by using point-cloud based co-occurrence texture methods, we now investigate the effect of lower vegetation, such as grass, in very flat marshland conditions. When vegetation is very low, and thus the effect very small, it appears difficult to separate the systematic shift from random measurement noise. When the shift becomes larger, however, it shows significant correlation with texture measures such as slope texture and standard deviation. These measures are derived from the laser data itself and do not require any additional information.