Elsevier

Information Sciences

Volume 243, 10 September 2013, Pages 1-19
Information Sciences

A new approach to create textured urban models through genetic algorithms

https://doi.org/10.1016/j.ins.2013.03.053Get rights and content

Abstract

The automatic process for texturing 3D urban environments realistically is a challenging topic in Computer Graphics, especially when the model is generated from 2D GIS data from real sites. In order to achieve realistic scenes, buildings, urban furniture and street roadways should be modeled automatically also considering the slope of steep streets. A texturing process by hand which takes into account all the urban elements of a city is an impracticable task, especially if quality results are required.

In this paper we propose an automatic method for building texturization in entire urban models using a small set of images. The different features of lower and upper parts of buildings are considered by implementing two different genetic algorithms. Our method can be used in cities with both horizontal and steep streets because the street slope is processed by the algorithm.

Introduction

3D City Modeling (3DCM) is a research area of great interest with applications in many different fields like games, tourism, and movie sets. When these models require the management of real data, the modeling process should also use the existing data sources, for example 2D urban GIS.

Currently there is great interest that these virtual environments, which may represent large cities, are accessed via the Internet. Additional requirements should be considered if pedestrian navigation is enabled, since the observer is very close to the urban elements and these should appear sufficiently realistic. In the case of sloping cities, the texturing process should consider additional restrictions in order to avoid improbable situations [14].

Since most city buildings are prism-shaped residential buildings, these urban elements are automatically modeled as 2.5D objects. Building texturization is essential for obtaining realistic results. This means that some problems found when the gate and frieze appear below ground level should be avoided (Fig. 1). Neither is realistic if the same texture is used in nearby buildings or if similar urban elements appear with very different sizes. For example, the size of windows should be in proportion with that of doors, as well as with the façades of the nearest buildings.

In this paper we propose an automatic method for texturing the buildings of an entire urban model in a client–server system. The 3D geometry of buildings and streets have been obtained according to the method described in [15] and depicted in Fig. 2. We use a set of real photographs as input data, which have been previously classified into several categories according to the place, where they were taken and considering the style and antiquity of these buildings. This new technique allows the automatic texturization of all blocks in the scene. The resulting scene does not exactly match with the real city because the same texture image can be assigned to several buildings. However, the models obtained are realistic enough since the whole environment has been texturized with real images and with all the architectural elements well placed on building façades.

The following section describes some previous work related to texturization in large urban models. Afterwards, in Section 3, we explain the treatment performed on the textures and the methodology for placing them. Next, in Section 4, we describe the main parameters in the genetic algorithms for obtaining the accurate results revealed in Section 5. Finally we provide the main conclusions and direction of future work.

Section snippets

Previous work

Rendering large-scale urban models is a challenging and complex task. It requires the generation of geometry and the texturization of all the elements in the scene. Procedural methods are generally used to this purpose, as can be found in the literature [13], [10], [11].

Other authors in several studies [17], [18], [24] build a textured mapped 3D model of each urban environment (a building, street roadway, etc.) from a set of real photographs. The problem with these techniques is the huge number

System overview

In this paper we propose a method for texturing 3D models of real buildings. Specifically, we create a 2.5D model from the footprint geometry of each building. The input data to this procedure are the 2.5D geometric models generated in [15] and a set of real photographs from the city being considered. These images must be previously processed and stored in a database. To make the texturization process easier, buildings have been divided into two different portions: top and ground floors, as

Genetic algorithm

The great amount and diversity of buildings makes texturization by hand an unfeasible task. We cannot manage all the façade images of a whole city. However the use of the same image for texturing several distant buildings can be accepted as valid. Thus, a reduced collection of images and a set of constraints can model a whole virtual urban environment. The method proposed in this paper performs a selection of valid images for texturing building façades according to the criteria of realism

Results

In this section we describe the results of our method from two points of view: performance and graphical results. Results are based on data from the city of Jaén (Spain) as mentioned in Section 4.1.

Conclusions and future work

In this paper we introduce a method for automatically texturizing the buildings of a whole city model using a reduced number of images. The buildings have been divided into two portions: top and ground. In each of them, a different genetic algorithm is implemented in order to select a set of valid textures. Regarding the results obtained, we can conclude that our method can be useful for texturing large urban environments over flat or hilly terrain.

In contrast to deterministic methods, genetic

Acknowledgements

This work has been partially granted by the Ministerio de Ciencia y Tecnología of Spain and the European Union by means of the ERDF funds, under the research project TIN2011-25259 and by the Conserjería de Innovación, Ciencia y Empresa of the Junta de Andalucía, under the research project P07-TIC-02773.

References (27)

  • Paul S. Heckbert, Fundamentals of texture mapping and image warping, Master’s thesis, CS Dept, UC Berkeley, May...
  • Z. Kang, Z. Zhang, J. Zhang S. Zlatanova, Geomatics Solutions for Disaster Management, chapter Rapidly realizing 3D...
  • Ailsa H. Land, Alison G. Doig, An automatic method for solving discrete programming problems, in: Michael Jnnger,...
  • Cited by (7)

    • A novel adaptive hybrid crossover operator for multiobjective evolutionary algorithm

      2016, Information Sciences
      Citation Excerpt :

      Based on the inspired evolutionary mechanisms, EAs can be classified into the following categories, such as genetic algorithms (GAs) [13,24], evolutionary strategies (ESs) [2,29], genetic programming (GP) [27,28], evolutionary programming (EP) [5,17], and other nature-inspired algorithms [4,7,18,51,52,56]. Among the existing EAs, GAs are widely recognized as the most popular and commonly used optimization algorithms, which have been applied in many engineering applications [30,48,49]. Naturally, as a kind of population-based random search approach, GAs mimic the natural selection principle “survival of the fittest” from the biological world.

    • Dense 3D reconstruction combining depth and RGB information

      2016, Neurocomputing
      Citation Excerpt :

      Experimental results show that the proposed method is effective. In the future research, we will try to extend the effective distance beyond 4 m and integrate more recent studies, including [15,25,28,30,38], to further improve the performance of this system. We will also try to implement the proposed method for specific applications [19,23,31–33,41,44,46,47,49,50], such as 3D object retrieval and mobile visual search, to further prove the validity of this method.

    • A genetic algorithm for task scheduling on heterogeneous computing systems using multiple priority queues

      2014, Information Sciences
      Citation Excerpt :

      A guided-random-search-based algorithm typically requires sufficient sampling of candidate solutions in the search space and has shown robust performance on a variety of scheduling problems. It is well known that particle swarm optimization (PSO) [13–17], ant colony optimization (ACO) [18,19], artificial bee colony algorithm (ABC) [20], simulated annealing (SA) [21,22], cuckoo search algorithm (CS) [23], tabu search (TS) [24,25], evolution algorithm [26,27], and genetic algorithms (GA) [24,28–42] have been successfully applied to the scheduling. GAs have been widely used to evolve solutions for many task scheduling problems.

    • Quaternion moment and its invariants for color object classification

      2014, Information Sciences
      Citation Excerpt :

      Texture is an important feature of objects in an image. Texture classification is a fundamental problem in computer vision and pattern recognition tasks [2,57,7,56,31,49,23,21,19,22,29]. The objective of the following experiments is to verify the performance of the proposed QMIs for the textured color image sets.

    • Efficient visibility determination in urban scenes considering terrain information

      2017, ACM Transactions on Spatial Algorithms and Systems
    View all citing articles on Scopus
    View full text