A Gestalt rules and graph-cut-based simplification framework for urban building models

doi:10.1016/j.jag.2014.09.012

International Journal of Applied Earth Observation and Geoinformation

Volume 35, Part B, March 2015, Pages 247-258

https://doi.org/10.1016/j.jag.2014.09.012 Get rights and content

Highlights

•
An optimization framework for clustering and generalizing large-scale urban building footprints and facade textures is presented.
•
The building footprints are partitioned into potential Gestalt groups.
•
The graph-cut-based optimization function is employed to obtain a consistent segmentation of the buildings.
•
An effective data structure is introduced to manage the aggregated building footprints and facade textures.

Abstract

To visualize large urban models efficiently, this paper presents a framework for generalizing urban building footprints and facade textures by using multiple Gestalt rules and a graph-cut-based energy function. First, an urban scene is divided into different blocks by main road networks. In each block, the building footprints are partitioned into potential Gestalt groups. A footprint may satisfy several Gestalt principles. We employ the graph-cut-based optimization function to obtain a consistent segmentation of the buildings into optimal Gestalt groups with minimal energy. The building footprints in each Gestalt group are aggregated into different levels of detail (LODs). Building facade textures are also abstracted and simplified into multiple LODs using the same approach as the building footprint simplification. An effective data structure termed SceneTree is introduced to manage these aggregated building footprints and facade textures. Combined with the parallelization scheme, the rendering efficiency of large-scale urban buildings is improved. Compared with other methods, our presented method can efficiently visualize large urban models and maintain the city's image.

Introduction

Real 3D city models have wide applications in urban planning, mapping and virtual tourism. With the rapid development of photogrammetry, computer vision, scanners and 3D modeling technologies, it is currently possible to construct detailed 3D city models in a practical and cost-efficient manner. This is possibly the main reason that 3D digital urban models have become so increasingly popular. These developments are changing the way the user conceives 3D data. Typically, a large city often contains hundreds of millions of buildings, and it is difficult to view digitized city models in real time. Therefore, the visualization of large-scale 3D city models for a variety of professional and mass-market services has received significant attention in the photogrammetry and computer graphics communities. To make such services appealing to a large audience, these 3D models should reach a sufficient level of realism and accuracy. Many solutions have been proposed to generate 3D models of huge urban environments (Royan et al., 2007). For example, Sheppard and Cizek (2009) proposed criteria for evaluating landscape visualizations under the categories of (i) accuracy, (ii) representativeness, (iii) visual clarity, (iv) interest, (v) legitimacy, (vi) access to visual information, and (vii) framing and presentation. To visualize a large-scale city efficiently, some generalization algorithms (such as Forberg, 2007, Zhang et al., 2012) are used to simplify urban building models.

To help people recognize a city more effectively and maintain spatial coherence, Gestalt principles are introduced to identify the distribution pattern of urban buildings. In this paper, we apply Gestalt rules and graph-cuts to cluster similar buildings into the same group. Then, the building footprints and façade textures are aggregated, and the whole urban models can be generalized into multiple levels. To obtain a good generalized representation of a city that abides by the Gestalt principles, the spatial relations, including distance, orientation, similarity and continuity, should be taken into account when building models are clustered. To reduce the load time, a multiple representation data structure termed SceneTree is proposed to store hierarchical models of the aggregated building footprints and facade textures. When these buildings are viewed, different levels of the building models are retrieved from SceneTree and rendered by the parallelization scheme. Compared with other methods, the novelty of our approach is that we implement an optimization framework for generalizing building footprints and facade textures simultaneously. In this framework, a graph-cut-based optimization method is employed to solve the conflicts that a footprint or facade texture may satisfy with several Gestalt principles, and SceneTree is created to store the generalized footprints and textures at deferent levels.

Section snippets

Related work

Numerous studies have developed a range of tools and techniques for visualizing city buildings. Additionally, an extensive amount of literature can be found in the related fields of computer vision, object recognition and geometric modeling. However, there remain many challenges in applying landscape visualization techniques for effectively communicating processes and changes in building landscapes. In this section, previous generalization results of 3D building models are summarized.

Methodology

This section presents our method for urban building generalization and visualization.

Experiments

To validate the performance of our method, we use VC++ 2010 and OpenGL to develop a 3D urban model visualization system. Experiments are performed on a Windows 7 Professional operating system with a 3.2 GHz Intel (R) Core (TM) i5-3470 CPU, 4 GB memory, and an Nvidia Geforce GT 620 graphics card. The study data comprise Beijing urban models with 122,056 buildings and 916,427 vertices.

Conclusion

According to the theory of Gestalt psychology, including proximity, similarity, continuity and common orientation (Wertheimer, 1923), separate elements in an image that have certain relations tend to be merged into a whole in people's mind. Accordingly, to obtain a good generalized representation of a city that abides by these criteria, the spatial relations, including distance, orientation, similarity and continuity, should be taken into account when building models are clustered. To speed up

References (29)

A. Forberg
Generalization of 3D building data based on a scale-space approach
ISPRS J. Photogramm. Remote Sens.
(2007)
B. Mao et al.
Generalization of 3D building texture using image compression and multiple representation data structure
ISPRS J. Photogramm. Remote Sens.
(2013)
B. Mao et al.
A multiple representation data structure for dynamic visualisation of generalised 3D city models
ISPRS J. Photogramm. Remote Sens.
(2011)
S.R. Sheppard et al.
The ethics of Google Earth: crossing thresholds from spatial data to landscape visualisation
J. Environ. Manag.
(2009)
J. Xie et al.
Automatic simplification and visualization of 3D urban building models
Int. J. Appl. Earth Observ. Geoinf.
(2012)
M. Zhang et al.
A geometry and texture coupled flexible generalization of urban building models
ISPRS J. Photogramm. Remote Sens.
(2012)
S. Ali et al.
Compressed facade displacement maps
IEEE Trans. Vis. Comput. Graph.
(2009)
C. Andújar et al.
Omni-directional Relief Impostors, Computer Graphics Forum
(2007)
C. Andújar et al.
Visualization of large-scale urban models through multi-level relief impostors
Comput. Graph. Forum
(2010)
C. Andujar et al.
Relief impostor selection for large scale urban rendering

Y. Boykov et al.

An experimental comparison of min-cut/max-flow algorithms for energy minimization in vision

IEEE Trans. Pattern Anal. Mach. Intell.

(2004)

Y. Boykov et al.

Fast approximate energy minimization via graph cuts

IEEE Trans. Pattern Anal. Mach. Intell.

(2001)

R. Chang et al.

Legible simplification of textured urban models

IEEE Comput. Graph. Appl.

(2008)

P. Cignoni et al.

Ray-casted blockmaps for large urban models visualization

Comput. Graph. Forum

(2007)

Cited by (22)

Second-order texton feature extraction and pattern recognition of building polygon cluster using CNN network
2024, International Journal of Applied Earth Observation and Geoinformation
The cluster patterns of features in map space represent a comprehensive reflection of individual feature geometric attributes and their spatial adjacency relationships. These patterns also embody spatial cognition results under the Gestalt principle. Describing non-linear spatial cluster patterns as effective regular structures is one of the fundamental tasks in deep learning for recognizing feature cluster patterns. In this study, based on the concept of texture co-occurrence matrices from regular gray-scale images, we utilized Voronoi diagrams to construct the tessellation structure of building polygons. Built upon the foundation of first-order texton co-occurrence matrices, we established three-dimensional texton co-occurrence matrices for building polygons, considered five attributes of building size, shape, orientation, and density, and encompassed 64 different combinations of second-order neighboring directions. This matrix concretizes the latent Gestalt spatial characteristics of building polygon clusters into a three-dimensional sparse matrix. It is then used as an input vector to construct a deep convolutional neural network for recognizing building polygon cluster patterns. Through adjustments and optimizations of neural network structure and strategies, along with validation through practical case studies and comparisons with other models, we have demonstrated the effectiveness of the second-order texton co-occurrence matrix in describing the characteristics of building polygon clusters.
An adaptive-size vector tile pyramid construction method considering spatial data distribution density characteristics
2024, Computers and Geosciences
Vector tile pyramid is a technology that provides a compact representation of geospatial data. It enables efficient transmission and rendering by storing geographic information in a tile-based format on the server side. Traditional vector tile construction methods divide vector data into a series of tiles of fixed size, such as 256*256 pixels, which leads to numerous empty tiles, imbalanced data distribution between tiles, and diminished internet transmission efficiency. To address these limitations, we propose a novel three-step method for dynamically adjusting tile sizes based on spatial data distribution density during construction. Firstly, we generalize the raw data into a multi-resolution vector dataset encompassing varying levels of detail (LOD). Subsequently, we employ a quadtree-based approach for each level within the multi-resolution dataset to construct adaptive-size vector tiles that align with the specific spatial distribution density characteristics. Finally, we encode these different-size vector tiles using Geohash technology to facilitate spatial indexing and internet transmission. Experimental results validate the effectiveness of our approach, including a notable reduction in the number of generated tiles, a more balanced distribution of data volume among them, and a marked enhancement in internet transmission efficiency.
Shape-preserving mesh decimation for 3D building modeling
2024, International Journal of Applied Earth Observation and Geoinformation
We propose a shape-preserving building model reconstruction method that involves simplifying the original building mesh to accommodate various building shapes and complexities. To achieve this, we apply a structure-aware segmentation technique to parse the ubiquitous building points into building geometric primitives and building structural points, i.e., anchor points. After that, we generate dense building meshes from building semantic points in a topology-aware manner. As the geometric primitive semantics are assigned to the building points during the structure-aware segmentation process, these primitive semantics of the building points can be explicitly transferred into the created building meshes. To offer lightweight and accurate building models with enriched semantics, we leverage the building structural points as constraints for the subsequent edge collapse simplification algorithm. This algorithm effectively decimates irrelevant vertices and meshes, while preserving the essential building structural contours. The entire simplification process is performed in a shape-preserving manner, granting us flexible control over the imposition of different degrees of strength regarding various geometric primitives during the simplification. We conduct qualitative and quantitative analyses to evaluate the effectiveness of our method on both individual building models and a large-scale urban scene. Additionally, we extensively compare our proposed shape-preserving algorithm with other state-of-the-art mesh decimation methods to demonstrate our superiority.
Long-term spatio-temporal changes of wetlands in Tibetan Plateau and their response to climate change
2023, International Journal of Applied Earth Observation and Geoinformation
Understanding the changes of wetlands on the Tibetan Plateau (TP) is important for action to ensure ecosystem resilience in Asia. However, mapping long-term changes of wetlands at high resolutions remains challenging. Here, we quantify the spatio-temporal changes of TP wetlands from 1990 to 2019, by combining Landsat imagery with deep learning to map TP wetlands. The deep learning model combined with transfer learning strategies achieves high classification performance using a few class samples. The validation results show that the user’s accuracy is 95.5% and the producer's accuracy is 90.1% for wetland extraction, satisfying with subsequent analysis of wetland spatio-temporal changes. Based on the wetland extraction model, we have created annual wetland map in the TP for the first time. We find that the areal extent of TP wetlands has increased by 31.2 ± 6.6 % over the past 30 years. The growth is particularly noticeable (by 22.5 ± 6.2 %) during 2015–2019. Spatially, the wetland areal extent on the Qiangtang Plateau (in the inner part of TP and as habitats of various birds and rare wild animals) and the source region of Yangtze River show the largest expansions by 55.3 ± 9.3 % and 44.0 ± 8.9 %, respectively. Such rapid wetland expansions are associated with increasing rainfall and temperature which have heterogeneous influences on wetland changes across the TP. Our findings provide evidence for the impact of climate change on wetland area. The marked wetland changes highlight that climate mitigation is a priority for high-latitude ecosystems.
Predicting residential building age from map data
2019, Computers, Environment and Urban Systems
The age of a building influences its form and fabric composition and this in turn is critical to inferring its energy performance. However, often this data is unknown. In this paper, we present a methodology to automatically identify the construction period of houses, for the purpose of urban energy modelling and simulation. We describe two major stages to achieving this – a per-building classification model and post-classification analysis to improve the accuracy of the class inferences. In the first stage, we extract measures of the morphology and neighbourhood characteristics from readily available topographic mapping, a high-resolution Digital Surface Model and statistical boundary data. These measures are then used as features within a random forest classifier to infer an age category for each building. We evaluate various predictive model combinations based on scenarios of available data, evaluating these using 5-fold cross-validation to train and tune the classifier hyper-parameters based on a sample of city properties. A separate sample estimated the best performing cross-validated model as achieving 77% accuracy. In the second stage, we improve the inferred per-building age classification (for a spatially contiguous neighbourhood test sample) through aggregating prediction probabilities using different methods of spatial reasoning. We report on three methods for achieving this based on adjacency relations, near neighbour graph analysis and graph-cuts label optimisation. We show that post-processing can improve the accuracy by up to 8 percentage points.
Recognition of building group patterns in topographic maps based on graph partitioning and random forest
2018, ISPRS Journal of Photogrammetry and Remote Sensing
Citation Excerpt :
Graph partition is to eliminate weak relationship among buildings within the connected subgraph and then forms a new disconnected graph that contains multiple connected subgraphs. To quantify the degree of the relationship among topographic objects such as buildings, metrics such as the centric point distance, minimum distance of building boundaries, mean distance, adjacent distance, and synthesized index, have been used (Regnauld, 2001; Ai and Zhang, 2007; Yang et al., 2011; W. Wang et al., 2015; Y. Wang et al., 2015). However, the relationship between each two objects within a group is also influenced by adjacent objects within the same group.
Recognition of building group patterns (i.e., the arrangement and form exhibited by a collection of buildings at a given mapping scale) is important to the understanding and modeling of geographic space and is hence essential to a wide range of downstream applications such as map generalization. Most of the existing methods develop rigid rules based on the topographic relationships between building pairs to identify building group patterns and thus their applications are often limited. This study proposes a method to identify a variety of building group patterns that allow for map generalization. The method first identifies building group patterns from potential building clusters based on a machine-learning algorithm and further partitions the building clusters with no recognized patterns based on the graph partitioning method. The proposed method is applied to the datasets of three cities that are representative of the complex urban environment in Southern China. Assessment of the results based on the reference data suggests that the proposed method is able to recognize both regular (e.g., the collinear, curvilinear, and rectangular patterns) and irregular (e.g., the L-shaped, H-shaped, and high-density patterns) building group patterns well, given that the correctness values are consistently nearly 90% and the completeness values are all above 91% for three study areas. The proposed method shows promises in automated recognition of building group patterns that allows for map generalization.

View all citing articles on Scopus

View full text

A Gestalt rules and graph-cut-based simplification framework for urban building models

Highlights

Abstract

Introduction

Section snippets

Related work

Methodology

Experiments

Conclusion

ISPRS J. Photogramm. Remote Sens.

ISPRS J. Photogramm. Remote Sens.

ISPRS J. Photogramm. Remote Sens.

J. Environ. Manag.

Int. J. Appl. Earth Observ. Geoinf.

ISPRS J. Photogramm. Remote Sens.

Compressed facade displacement maps

IEEE Trans. Vis. Comput. Graph.

Omni-directional Relief Impostors, Computer Graphics Forum

Visualization of large-scale urban models through multi-level relief impostors

Comput. Graph. Forum

Relief impostor selection for large scale urban rendering

An experimental comparison of min-cut/max-flow algorithms for energy minimization in vision

IEEE Trans. Pattern Anal. Mach. Intell.

Fast approximate energy minimization via graph cuts

IEEE Trans. Pattern Anal. Mach. Intell.

Legible simplification of textured urban models

IEEE Comput. Graph. Appl.

Ray-casted blockmaps for large urban models visualization

Comput. Graph. Forum