Titre : |
Acquisition of 3D topography : automated 3D road and building reconstruction using airborne laser scanner data and topographic maps |
Type de document : |
Thèse/HDR |
Auteurs : |
Sander J. Oude Elberink, Auteur |
Editeur : |
Delft : Netherlands Geodetic Commission NGC |
Année de publication : |
2010 |
Collection : |
Netherlands Geodetic Commission Publications on Geodesy, ISSN 0165-1706 num. 74 ![](./images/globe.gif) |
Importance : |
172 p. |
Format : |
17 x 24 cm |
ISBN/ISSN/EAN : |
978-90-6132-318-1 |
Note générale : |
Bibliographie |
Langues : |
Anglais (eng) |
Descripteur : |
[Vedettes matières IGN] Lasergrammétrie [Termes IGN] bati [Termes IGN] carte topographique [Termes IGN] données laser [Termes IGN] données localisées 3D [Termes IGN] lasergrammétrie [Termes IGN] modèle 3D de l'espace urbain [Termes IGN] reconstruction 3D [Termes IGN] reconstruction 3D du bâti [Termes IGN] réseau routier [Termes IGN] télémétrie laser aéroporté
|
Index. décimale : |
33.80 Lasergrammétrie |
Résumé : |
(Auteur) Introduction and research goal : Our research covers the automation in acquiring three dimensional (3D) topographic objects. The research tasks focus on two specific objects: roads and buildings. These objects are of high importance in 3D city models as they are two major topographic classes in the urban environment. Our activities are located between: -1. how topographic objects exist in reality; -2. how they are captured in the data, and -3. how they appear in a modelled/virtual world. To accomplish an automated approach, existing 2D topographic maps are upgraded to 3D using airborne laser scanner data. 3D topography also includes multiple heights or even multiple objects on top of each other at a certain location. The essence in the research activities on roads differs basically from those on buildings. For roads the focus is on reconstructing the edges' height of the objects, whereas for buildings the challenge is to reconstruct the 3D polyhedral roof shape inside the building edges.
3D Road reconstruction : When examining 3D road objects, we can expect that multiple road objects cross at a certain location. An automated method for 3D modelling of complex highway interchanges is presented. Laser data and 2D topographic map data are combined in an innovative 3D reconstruction procedure. Complex situations demand for knowledge to guide the automatic reconstruction. This knowledge is used in the fusion procedure to constrain the topological and geometrical properties of the reconstructed 3D model. Laser data has been segmented and filtered before it is fused with map data. In the surface-growing algorithm combining map and laser points, the laser data is assigned to the corresponding road element. Elevations of map points are determined by least squares plane fitting through a selection of neighbouring laser points. Although results are shown using two specific data sources, the algorithm is designed to be capable of dealing with any polygon-based topographic map and any aerial laser scanner data set. Quality analysis is essential for developing a reliable reconstruction process and for a proper use of 3D data. The quality of 3D reconstructed roads strongly depends on accuracy and type of input data and the reconstruction processing steps. We predict the precision of reconstructed map elevations by propagating errors in the input data through the processing steps. Besides this quality prediction, we test the reconstructed model against independent reference data. Differences between these two datasets are explained by the predicted uncertainty in the model. Map point heights can be reconstructed with an average precision of 10 to 15 cm, depending on the laser point configuration.
3D. Building reconstruction : The building reconstruction task contains three main goals: -1. to select laser points belonging to building roofs, -2. to detect the roof structure of that building, and -3. to reconstruct the outlines of the roof. We present a building reconstruction approach, which is based on a target graph matching algorithm as intermediate step to relate laser data with building models. Establishing this relation is important for adding building knowledge to the data. Our targets are topological representations of the most common roof structures which are stored in a database. Laser data is segmented into planar patches. The segments that are selected in the segment-in-polygon algorithm are considered initial roof segments. Topological relations between segments, in terms of intersection lines and height jumps, are represented in a building roof graph. These relations are labelled according to their geometry and that of the segments (e.g. same/opposite normal direction, convex/concave, tilted/horizontal). This graph is matched with the graphs from the target database. Matching results describe which target objects appear topologically in the data. Our target based graph matching algorithm supports the first two goals. The matching algorithm performs a filtering task: data features that topologically correspond with common roof structures are considered to be part of the roof structure of that building. These data features will be transferred to our automated building reconstruction, where the outlines of the roof faces have to be reconstructed. Segments and intersection lines that do not fit to an existing target roof topology will be removed from the further automated reconstruction approach. The reconstruction algorithm covers the third main goal of our building reconstruction task. For the geometric reconstruction, we present two approaches that vary in the amount of information they take from the data. The first, more data driven approach starts with laser data features that have been matched with target models. In general, the matched intersection lines represent the interior of the roof structure, so the task is to find an appropriate solution for the remaining roof edges, e.g. eaves and gutters. Map data is used for selection of roof segments and is taken as location for walls. Therefore we need to split up map polygons in order to build walls that distinguish various height levels, e.g. at step edge locations. The second, more model driven approach reconstructs parameterised building models. This approach relies more on geometric assumptions, such as roof symmetry, but the models can be refined if the data deviates significantly from the model. The target information includes the details on how these deviations are determined and on the thresholds to decide what is significant or not. We present results of 3D reconstructed models, including several quality checks. These quality measures describe the completeness of the match results plus the correctness of assumptions to the roof outline. About 20% of the buildings are affected by segments that did not completely match with the target graphs. In a few of these cases, this is correct because the segment is not representing a roof face. However, in about 40% of these cases, a neighbouring segment that would complete a target match is missing. Adapting processing parameters, such as minimum segment size, may improve the result but it may also disturb other topological relations. Setting the parameters is therefore an important task for the operator. Specially, parameters that define the segmentation algorithm are crucial as the segment is the key data feature in our building reconstruction algorithm. In order to improve our matching algorithm, the likelihood of relations between segments could be included in the attribute list of edges in the roof topology graph. At the moment only information on the geometric appearance of the intersection line is given as attribute value to the corresponding graph edge. Future work includes defining likelihood functions for graph edges and analysing the effect of likelihood attributes. |
Note de contenu : |
Part 1: Introduction to acquisition of 3D topography
1 Introduction
1.1 3D Topography
1.2 Scope and limitations
1.3 Input data
1.4 Research problems
1.5 Goal and objectives
1.6 Importance
1.7 Thesis outline
2 Use of 3D topography
2.1 Introduction
2.2 User requirements
2.2.1 Municipality of Den Bosch
2.2.2 Survey Department of Rijkswaterstaat
2.2.3 Water board "Hoogheemraadschap de Stichtsche Rijnlanden"
2.2.4 Topographic Service of the Dutch Cadastre
2.3 Re-using 3D models
2.3.1 Municipality of Den Bosch
2.3.2 Survey Department of Rijkswaterstaat
2.3.3 Water board "Hoogheemraadschap de Stichtsche Rijnlanden"
2.3.4 Topographic Service of the Dutch Cadastre
2.3.5 Availability and distribution
2.3.6 Data fusion
2.3.7 Generalization and filtering
2.3.8 3D Represents as-is situation
2.4 Role of use cases in research project
2.5 Recent developments in using 3D topography
2.6 Conclusions
Part 2: 3D Roads
3 3D Reconstruction of roads
3.1 Introduction
3.2 Related work
3.2.1 Road reconstruction from aerial images
3.2.2 2D Road mapping from laser data
3.2.3 3D Reconstruction from laser data
3.3 Proposed approach
3.4 Data sources
3.4.1 Airborne laser scanner data
3.4.2 Pre-processing laser data
3.4.3 2D Topographic map data
3.4.4 Pre-processing 2D map
3.5 Fusion of map and laser data
3.5.1 Research problems on fusing map and laser data
3.5.2 Proposed fusion algorithm
3.6 3D Reconstruction of polygons
3.6.1 Polygon boundaries
3.6.2 Additional polygons
3.6.3 Assumptions on boundaries
3.6.4 Surfaces
3.7 Results
3.7.1 Interchange "Prins Clausplein"
3.7.2 Interchange "Waterberg"
3.8 Discussion
3.8.1 Parameter settings
3.8.2 Topological correctness
4 Quality analysis on 3D roads
4.1 Error propagation
4.1.1 Quality of plane at map point location
4.1.2 Quality of laser block
4.1.3 Quality of plane model
4.2 Reference data
4.2.1 Height differences between reference data and 3D model
4.3 Testing of predicted quality
4.4 Discussion
Part 3: 3D Buildings
5 Building shape detection
5.1 Introduction
5.1.1 Real buildings vs 3D model representation
5.1.2 Real buildings vs appearance in input data
5.1.3 Appearance in input data vs 3D model representation
5.2 Related work
5.2.1 2D Mapping of building outlines
5.2.2 3D Reconstruction of buildings
5.3 Research problems
5.3.1 Problems on roof shape detection
5.3.2 Problems on scene complexity
5.4 Proposed approach
5.5 Information from map data
5.6 Features from laser data
5.6.1 Segmentation of laser scanner data
5.6.2 Intersection lines
5.6.3 Step edges
5.6.4 Roof topology graph
5.7 Target graphs
5.8 Target based graph matching
5.9 Complete matching results
5.10 Incomplete matching results
6 3D Building Reconstruction
6.1 Introduction
6.2 Components of a roof boundary
6.3 Approach 1: Combine features from complete match results
6.4 Extension of horizontal intersection lines
6.5 Outer boundaries of roof faces
6.5.1 Flat roof faces
6.5.2 Eave construction
6.5.3 Gutter construction
6.6 Dormers and step edges
6.6.1 Simple dormers
6.6.2 Step edges
6.6.3 Step edges for map subdivision
6.7 Reconstruction of walls
6.8 Approach 2: reconstructed targets
6.8.1 Parameterised target models
6.8.2 Use of map data
6.8.3 Limitations
6.8.4 Potential use
6.9 Summary
7 Results and evaluation
7.1 Introduction
7.2 Results
7.2.1 Approach 1: Combined features
7.2.2 Approach 2: Reconstructed targets
7.3 Evaluation
7.3.1 Laser data features
7.3.2 Evaluation on target based matching
7.3.3 Reconstructed models
7.3.4 Problematic situations
7.3.5 Performance in time
7.4 Potential for nation wide 3D building database
7.5 Summary
Part 4: Conclusions and recommendations
8 Conclusions and recommendations
8.1 Conclusions
8.1.1 3D Topographic object reconstruction
8.1.2 3D Road reconstruction
8.1.3 3D Building reconstruction
8.2 Recommendations |
Numéro de notice : |
10833 |
Affiliation des auteurs : |
non IGN |
Thématique : |
IMAGERIE |
Nature : |
Thèse étrangère |
DOI : |
sans |
En ligne : |
https://www.ncgeo.nl/index.php/en/publicatiesgb/publications-on-geodesy/item/258 [...] |
Format de la ressource électronique : |
URL |
Permalink : |
https://documentation.ensg.eu/index.php?lvl=notice_display&id=62510 |
| ![Acquisition of 3D topography vignette](https://documentation.ensg.eu/img_cache/pmb_ign_record_62510.webp) |