A Comprehensive Automated 3D Approach for Building Extraction, Reconstruction, and Regularization from Airborne Laser Scanning Point Clouds

Abstract

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.

Keywords: building modeling; building outline; planar faces; regularization; segmentation.

Figure 1.

Comparison of measures d defined in 2.5D and 3D and applied to differently shaped areas of interest Γ (box and sphere). For two planes, A and B, inclined by 1^◦) and rotated around the y-axis, d is shown with respect to the inclination angle of A.

Figure 2.

Seed cluster determination by sequential evaluation of the individual plane parameters in the feature space. (a) shows the histograms of all points (most dominant parameter: a₁). The selected points (green) are analyzed further ((b) and (c)). Finally, 27 points define the seed cluster (d).

Figure 3.

Detection of planar faces from a point cloud by segmentation. (a)-(d): Determination of the segments 1 and 2. (a) and (c): Seed cluster points (large red dots), points accepted in object space (orange), and points accepted in feature space (green). (b) and (d): Result of the robust plane fit (dark green: accepted; red: rejected). The small cyan dots in (c) and (d) represent points already assigned to segment 1. Black points have not been used so far. (e): Final segmentation including roof and wall faces.

Figure 4.

Building outline determination initiated by mean shift segmentation (a) and (c) and planar face extraction (b). The points assigned to one building are shown as red circles in (c) and as magenta crosses in (d) and (f). The generalization of the 2D α-shape (cyan polygon in (d)) is applied using an angular criterion (e). The generalized (green) and the regularized outline (black) are shown in (f).

Figure 5.

Suggested workflow from building extraction to building modeling

Figure 6.

Different representations of 3D models of buildings with complex roof structures. (a): Waterproof model and color coded segmentation result; (b): Edge-model (red: eaves, green: gables, yellow: other edges); (c): 3D model with roof overhangs; (d): Roof model regularization.

Figure 7.

Result of building outline extraction. Building region estimation (green), extracted building outline (black), and original points (color: local aspect angle, gray: horizontal).

Figure 8.

Left: Analysis of building outline extraction (green: coarse estimation of the building region, black: extracted building outline, red: reference data). Right: Distances of original points with respect to the regression planes representing the roof segments.

Figure 9.

Comparison of parameters determined for individual buildings from the ALS point cloud and from independent reference data. Left: Differences of eave height, building height, building area (square root), and building perimeter are shown for each building (Difference: Reference minus ALS). Right: Box plots representing the same values.

Figure 10.

Left: Rendered city model correctly intersected with a DTM triangulation. Right: 3D city model with manually applied textures from terrestrial and airborne images.

Figure 11.

Model of a historical city determined by the proposed, data driven approach. The roofscape is shown in red and vertical walls in gray.

Figure 12.

Visual comparison of city models which were automatically determined from the same data set (cf. Figure 11), but by applying two different approaches (Left: proposed data driven approach; Right: model driven approach). Distances of the original points with respect to the planes defining the roof segments are shown color coded.