^{*}

This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (

High resolution Digital Surface Models (DSMs) produced from airborne laser-scanning or stereo satellite images provide a very useful source of information for automated 3D building reconstruction. In this paper an investigation is reported about extraction of 3D building models from high resolution DSMs and orthorectified images produced from Worldview-2 stereo satellite imagery. The focus is on the generation of 3D models of parametric building roofs, which is the basis for creating Level Of Detail 2 (LOD2) according to the CityGML standard. In particular the building blocks containing several connected buildings with tilted roofs are investigated and the potentials and limitations of the modeling approach are discussed. The edge information extracted from orthorectified image has been employed as additional source of information in 3D reconstruction algorithm. A model driven approach based on the analysis of the 3D points of DSMs in a 2D projection plane is proposed. Accordingly, a building block is divided into smaller parts according to the direction and number of existing ridge lines for parametric building reconstruction. The 3D model is derived for each building part, and finally, a complete parametric model is formed by merging the 3D models of the individual building parts and adjusting the nodes after the merging step. For the remaining building parts that do not contain ridge lines, a prismatic model using polygon approximation of the corresponding boundary pixels is derived and merged to the parametric models to shape the final model of the building. A qualitative and quantitative assessment of the proposed method for the automatic reconstruction of buildings with parametric roofs is then provided by comparing the final model with the existing surface model as well as some field measurements.

Automatic building reconstruction from Digital Surface Models (DSMs) with or without using other data sources is still an active research area in Photogrammetry and GIS institutions. In this context, providing a 3D CAD model that represents the overall shape of the building and containing the most significant parts has boosted many applications in the GIS area such as urban planning. In the past few years, several algorithms have been proposed for automated 3D building reconstruction. The algorithms comprise methods that only employ elevation data such as high resolution airborne LIDAR for model generation while some methods use other additional sources of data. An additional data source plus DSM is usually employed when the quality or resolution of the elevation data is not appropriate for model generation. Segmentation based approaches for a 3D building model generation from grid data are proposed by [

Rotensteiner [

Khoshelham [

A projection based approach for 3D model generation of the buildings from high resolution airborne LIDAR data was proposed by [

Kada [

Kabolizadeh [

In this paper we propose a method that aims at simplifying the 3D reconstruction of the building blocks by decomposing the overall model into several smaller ones corresponding to each building part. The focus is on the generation of 3D models of parametric building roofs as the basis for creating Level of Detail 2 (LOD2) according to City Geography Markup Language (CityGML) standard [

A similar method has been already reported [

The novelty of the approach proposed in this paper in comparison with [

The process begins with the extraction of building ridge lines using orthoimage and height information. According to each ridge line, a projection-based algorithm is employed to transfer the 3D points into 2D space by orthogonal projection of the corresponding pixels of each building part onto a 2D plane that is defined based on the orientation of the ridge line. Based on the type of the roofs, a predefined 2D model is fitted to the data, and in the next step, the 2D model is extended to 3D by analyzing the third dimension of the points. A final model regarding the parametric roof structures of the building block is defined by merging all the individual models and employing some post-processing refinements regarding the coinciding nodes and corners to shape the appropriate model. Additionally prismatic models with flat roof are provided regarding to the remaining areas that are not containing ridge lines. Finally, all parametric and prismatic models are merged to form a final 3D model of the building.

In this section, a new method is proposed for the reconstruction of buildings by integrating DSMs and orthorectified image information produced from Worldview-2 stereo satellite imagery. Worldview-2 provides panchromatic images with 50 cm Ground Sampling Distance (GSD) as well as eight-band multispectral images with 2 m GSD. A DSM is produced from panchromatic Worldview-2 images with 50 cm image resolution using a fully automated method [

In this article, the main focus is on the reconstruction of the buildings with tilted roof shapes, including hipped and gabled parametric roof models, although modeling of the flat roof segments (prismatic models), which is reported in a former work of the authors, is utilized in this study as well.

The preliminary and also the most critical issue on all the building reconstruction approaches is building segmentation, which is still an unsolved problem particularly based on only geometric data such as DSMs. During the past year, many algorithms have been proposed for building classification and segmentation from remotely sensed data. Some of these algorithms work properly for some special areas and fail in others. In this paper we do not focus on building segmentation issue, and therefore we assume that the building masks have been extracted using a successful algorithm with any degree of automation.

Accordingly, the automatic 3D building reconstruction algorithm proposed in this paper comprises the following major steps:

Ridge-based decomposition of building parts

Projection-based reconstruction of parametric roofs

Prismatic model generation related to flat roof segments

Merge parametric and prismatic models and refine the corner nodes

The idea of the 3D building reconstruction algorithm proposed in this article is to simplify the modeling procedure by decomposing the overall building model into smaller tiles based on the location of the ridge lines. The building parts containing ridge lines are individually modeled and merged to shape the overall model of the entire building. Accordingly, the location of the ridge lines in buildings with tilted roof structures should be carefully extracted. The quality of the final model has a direct relation to the quality of extracted ridge lines,

Ridge lines are the basis for decomposing a building block into smaller tiles.

Ridge lines are the basis for projection-based model generation of each part.

Therefore, the first and the most important and sensitive part of the proposed approach is extracting ridge lines. Arefi [

Surface normals of DSM: The surface normal is a vector perpendicular to a surface, which represents the orientation of a surface. It can be estimated by determining the best fitting plane over a small neighborhood. A normal vector can also be computed by means of the cross product of any two non-collinear vectors that are tangent to the surface at a targeted pixel [

Regional maxima from DSM: Here, an algorithm based on image reconstruction using geodesic morphological dilation [

Accordingly, geodesic dilation (δ_{I}

_{I}

Considering the reconstructed image of the example depicted in

Canny edges from orthorectified image:

The above mentioned three feature descriptors are employed to classify and extract the potential ridge points using DSM and orthorectified image as follows:

The image provided by surface normal (

Intersection of the three binary images,

Next, the RANdom SAmple Consensus (RANSAC) algorithm [

In this paper, it is assumed that an individual building part exists according to each ridge line. Therefore, for each ridge line and the pixels located in its buffer zone, a 3D model is fitted. For a rectangle buffer zone parallel (or perpendicular) to the main orientation, the points located inside it are extracted using the point-in-polygon algorithm. This step is necessary for buildings containing more than one part. A rectangle parallel to the main orientation (parallel to ridge line) is created. A rectangle is defined around the ridge line with equal distances of the edges to the ridge line. The limits of the rectangle are selected in this way that detected building pixels are all included. In

The procedure continues by projecting the localized points onto a 2D plane perpendicular to the ridge direction (

In order to shape the final 3D model relating to the building part, the 2D model is converted back to 3D space by extruding it orthogonally to the projection plane. The 3D model consists of four walls plus one to four roof planes: two inclined planes in addition to two vertical triangular planes for a gable roof, and four inclined planes for a hipped roof (

After reconstructing 3D models for all building parts, they are merged to form the overall 3D model of the building.

The method contains some extra processes to refine the nodes and the walls that represent common locations. In this context, the following refinements are applied (

If neighboring segments are nearly parallel and close to each other, the parallel sides are shifted to a location with equal distance to each other.

If the end point of a ridge line is close to the end point of the other ridge line, the intersection of two lines is replaced to both end points.

Two algorithms are proposed for the approximation of the building polygons based on the main orientation of the buildings [

If the building is formed by a rectilinear polygon,

If the building is not rectilinear,

RANSAC was originally devised to robustly fit one single model to noisy data. It turns out, however, that it can also be successfully used to fit a beforehand unknown number of models to the data: In the case of the ground plan boundaries, the number of line segments is initially unknown. We simply apply the method repeatedly—always deleting the already fitted given points from the input data—until either:

we consider the lines found so far sufficient to construct the ground plan completely, or

the number of points fitting to the best line segment with respect to the current iteration step falls below a chosen threshold

As an alternative to the RANSAC-based approximation algorithm, a method similar to MBR-based is proposed for the buildings containing several orientation directions. The method called Combined Minimum Bounding Rectangle (CMBR) based algorithm for hierarchical approximation of non-rectangular polygons. In this method, based on each orientation, a MBR (rectangle) polygon is estimated as first approximation level, as shown in

All approximation methods are containing the parameters as “stopping criteria”, which could be tuned to provide desired level-of-detail for 2D approximation. The MBR- and CMBR-based methods are both iterative algorithms and in each iteration the remaining regions are checked if they have sufficient number of pixels for approximation or they can be neglected.

In order to include the other structures (here, with flat roof) into the merged parametric model generated in Section 2.2, the ground plan of the merged model is compared with an approximated polygon. The overall area of the approximated polygon is subtracted from the corresponding area of the parametric models. The positive pixels belong to protrusions and the negative pixels are related to indentations. The areas corresponding to the protrusions and indentations are again approximated. The average of the heights of the internal points of protrusion area is used as the height of the building part. Although, this does not mean that the protrusion parts have always flat roofs, but since their corresponding roof types cannot be distinguished by the proposed algorithm, a prismatic model is fitted to the points.

In

A final model of the building block is provided by including the prismatic model corresponding to the protrusion area to the parametric models and excluding the indentation area from it. The corresponding polygon nodes of indentation and protrusion regions are included in the overall 3D model. Finally, the inclinations of the building roofs are adapted after including the indentation nodes.

The proposed algorithm for the 3D reconstruction of buildings from Worldview-2 DSM by integrating image information has been tested in an area located at the city center of Munich, Germany. The area contains 7 buildings with different shapes that are all modeled using the projection-based approach.

Additionally the comparison can be extended in 3D by superimposing the representation of the parametric models on a 3D surface generated from the DSM (

Accordingly, the quality of the model can be evaluated by the rate of visible colors against gray (height) pixels. In areas where the green colors are visible, the produced roof model is higher than the height pixels in the DSM. In contrast, the visible gray pixels on the roofs show that the roof model is located below the DSM in those areas. A similar conclusion describes the quality of the walls against DSM pixels.

In addition to visual interpretation and comparisons, a quantitative assessment proves the high quality of the fitted models. In order to measure the statistical parameters, the corresponding heights of the walls and ridges are taken from official construction plans of the buildings.

It is necessary to mention that due to the lack of a ground truth related to the horizontal coordinates (

An algorithm for the automatic 3D reconstruction of buildings using both the Worldview-2 DSM and the edge information from orthorectified images is proposed. According to the ridge information, the building block is decomposed into several parts depending on the number of ridge lines. For each ridge, a projection plane is defined and all points located on the buffer zone of the ridge line are projected onto that plane. Next, a 2D model supported by maximum number of projected points is modeled and then extended to 3D to shape a hipped- or gabled-roofs (parametric model). Integrating all 3D models corresponding to each ridge line produces the parametric model of the building block.

Additionally, prismatic models with flat roofs are provided regarding the remaining areas of the buildings, which are not already modeled by the projection-based method. Finally, all parametric and prismatic models are merged to form the final 3D models of the buildings. Here it is important to mention that all parameters, such as thresholds for binarization steps, thresholds for RANSAC line fitting, size and area thresholds, and other parameters, are empirically selected and defined in the program, which can be almost used for other Worldview DSM and orthorectified images with same resolution.

The example used in the previous section to illustrate the developed algorithm shows that the concept for building reconstruction provides an appropriate result in particular about the buildings isolated from the neighboring trees. Since the data used are only satellite data, it also shows that it is possible to derive already LOD2 building models from these data, which has a lower resolution in comparison with airborne data. It follows that for areas where no airborne measurements are available, satellite data could serve as the source for city modeling. Further applications could be 3D change analysis or damage assessment after earthquakes or other hazards.

However, from the quantitative assessment provided by

Boundary approximation using MBR and CMBR methods are efficient and robust techniques that employ rectangular and non-rectangular boundary polygons. In a future work we aim to expand the algorithm by the approximation of more complex building shapes such as buildings boundaries containing curves and lines.

Automatic ridge line extraction method using the combination of image and DSM data is an appropriate method and provides high quality ridge line with standard deviation of about 50 cm.

Detecting the building walls is more critical than detecting the ridges, particularly when the building is connected to a tree or the building wall is smoothed due to spatial interpolation after image matching. Nevertheless, if the buildings are isolated from the neighboring buildings or other 3D objects, the building wall could be extracted with a proper quality as seen in

Here it should be mentioned that the existence of the 3D model of roof superstructures containing tilted faces (such as attic windows) depends on whether the corresponding ridge line is extracted or not. As mentioned, if the corresponding ridge line contains enough number of pixels (depending on the image resolution and size of the ridge line), the line can be extracted by the RANSAC algorithm. This happens usually for very big castles containing sufficient number of pixels for a roof superstructure. A similar reason exists about other small structures such as chimneys, whether they contain enough number of pixels (area threshold) to be modeled or not.

Compared with previous methods, the strength of the proposed algorithm include: (a) it is a model-driven approach,

On the other hand, the drawback of the algorithm is its sensitivity to the location and accuracy of the extracted ridge lines. The quality of the final model is directly related to the quality of the extracted ridge line. Other limitation concerns the buildings connected to a tree from one side or surrounded by trees, which makes wall detection inaccurate.

The authors would like to thank Pablo d’Angelo for generating the DSM from Worldview-2 stereo data. We would also like to thank European Space Imaging (EUSI) for providing the Worldview-2 stereo data for scientific use.

Building model representation for LOD1 to LOD4 according to CityGML (taken from [

Ortho-rectified Worldview-2 (left column) versus DSM produced from Worldview-2 stereo satellite images (right column) Munich.

Workflow for projection based 3D building reconstruction.

Feature extraction from DSM and orthorectified images.

Applying geodesic reconstruction to extract the top pixels of a sample building.

Ridge extraction.

Projection-based model generation.

Final model is overlaid on

MBR-based polygon approximation.

Approximation of polygon obtained using RANSAC.

CMBR-based polygon approximation.

DSM—New Castle Stuttgart.

LOD1—Prismatic Model of New Castle Stuttgart.

Generating the final 3D model of a building containing parametric and prismatic roof structures.

Automatically generated 3D building models superimposed on

Final representation of

Ground Truth (GT)

1 | 15.00 m | 14.88 m | 20.40 m | 20.06 m |

2 | 15.00 m | 14.48 m | 20.40 m | 20.00 m |

3 | 14.28 m | 13.65 m | 22.78 m | 23.57 m |

4 | 15.00 m | 14.33 m | 20.40 m | 19.83 m |

5 | 11.15 m | 11.30 m | 14.25 m | 14.50 m |

6 | 11.15 m | 10.80 m | 14.25 m | 14.20 m |

7 | 11.15 m | 10.20 m | 14.25 m | 14.20 m |

| ||||

Mean | ||||

Std. Dev. |