3D Indoor Models and Their Applications
- 175 Downloads
KeywordsPoint Cloud Building Information Model Structure From Motion Indoor Space Shape Grammar
Indoor environments are often referred as to enclosed spaces. However, the general definition of space can already indicate that a space can be bounded. Wordnet (http://wordnet.priceton.edu) defines space as “an empty areas usually bounded in some way between things.” Specialized ontologies such as OmniClass (http://www.omniclass.org, a classification for architecture, engineering, and construction in North America) distinguish between spaces by form and spaces by function. “Spaces by form are basic units of the built environment delineated by physical or abstract boundaries and characterized by physical form.” “Spaces by function are basic units of the built environment delineated by physical or abstract boundaries and characterized by their function.” The spaces can be both 2D and 3D. For example, space by form can be a 3D room or a 2D walking path. An interesting example is a wall (interior, exterior), which is considered a space by function, which implies that spaces can be filled with some material, i.e., not just air.
Indoor spaces are artificial constructs designed and developed to support human activities. 3D indoor models, being a virtual digital representations of indoor spaces, have to be able to support these activities.
Approaches and Methodologies of Automatic Indoor Model Generation
Method(s) to be utilized
ID tags CAD files Documents
CAD/GIS files Point
BIM city models
BIM city models
Computer-aided design (CAD) and lately the architecture, engineering, and construction (AEC) are the oldest domain offering 3D tools for representation of indoors. CAD was primarily developed for engineers responsible for designing and building facilities (Azri et al., 2012). It is easy to compute and design with CAD tools due to its friendly environment and dynamic interaction. CAD tools which were dealing with large-scale and detailed models did not focus on maintenance of attributes and lack the support of geodetic reference systems. Although CAD models offer a convenience in representing indoor information, several drawbacks of CAD models have been revealed. For instance, CAD is only a platform to design and model geometries. Thus, information such as attribute, topology can only be tagged externally during the design process. Some new extensions of CAD/AEC (Bentley Systems, Autodesk products) do allow the maintenance of topology and semantics but in a quite vendor-dependent way. Therefore, the topology and semantics are lost when the model is exported to another software tool. If the information attached to the model is not transferred together with the model, the users can only interpret information from what they have seen through the model. In addition, if the building model was developed with low level of detail, there may not be much geometric and semantic information that can be extracted and used.
Building information model (BIM) is the next stage in the digital representation of building interiors and facilities. BIMs can be used to model building information in 3D with the support of an intelligent database that contains information for design decision making, production of accurate construction documents, prediction of performance factors, cost estimating, design scenario planning, and construction planning. BIM is an object-oriented, semantically rich model. The spatial relationships between building elements are maintained in hierarchical manner. It maintains many geometric primitives ranging from simple B-reps to free form curves and surfaces. Today, the most prominent BIM standard is the Industry Foundation Classes (IFC).
3D indoor models are investigated by researchers in GIS domain as well. Digital city models have become widely used for digital representation of major cities. With the advent of 3D city models such as in Google Earth, CityGML, and others, indoor modeling became a priority topic of research in GIS society. Today, CityGML is the best known model for 3D indoor modeling. CityGML is developed for representing 3D city geometry, (a kind of) topology, and thematic-semantic modeling. CityGML can be used to represent buildings and building parts and properties in different levels of detail (LOD) (i.e., from LOD0 up to LOD4). CityGML LOD4 provides a semantic-thematic model for representing indoors. The indoor objects are much less than the objects that can be represented in IFC. However, their simplicity seems quite sufficient for a large group of outdoor and indoor applications (Billen et al., 2014).
Which of the two most prominent standards will be used for 3D indoor modeling depends very much on the application. CAD/BIM domain has been traditionally dealing with very large-scale representations, while GIS with very small scale (up to km). In the last decade a fusion and overlap between the two domains is observed (Fig. 3). However, there are fundamental difference between the two models related to the conceptual definition of the indoor objects. IFC objects are defined from the view of the constructor and the CityGML LOD4 from the view of the user (Fig. 3). IFC is very appropriate to maintain information about construction parts of building as concrete walls, slabs, and columns. CityGML is focused on the modeling of the visible environment such as surfaces of the walls as part of one room or surfaces of walls as part of the façade of a building. This poses numerous challenges to the transformations between the two models (Fig. 4).
Indoor applications have been traditionally not a topic of research of GIS community. Designers, constructors, and engineers have been worked and used 3D indoor representations for modeling airflow simulation, smoke modeling, interior design, and facility management. However, the two prominent indoor applications are indoor navigation and facility management.
Generally speaking, a navigation system consists of the following components: positioning of a user, calculation of a best path (cheapest, fastest, safest, etc.) to some destination(s), and guidance along the path. Indoor navigation is a very prominent and active research area. It has been originated from navigation robots and it moved to human navigation in the last two decades. However, it remains a challenging topic for several reasons: indoor positioning is not very accurate, users can freely move within the building, topology model (or path network) construction process may not be straightforward due to complexity of indoor space, and humans need an appropriate guidance. Many papers have provided extended overview on navigation systems and models (2D and 3D) to support indoor navigation (Afyouni et al., 2012; Montello, 1993; Fallah et al., 2013; Bandi and Thalmann, 1998; Zlatanova et al., 2014). The majority of the indoor models found in current literature are still mostly 2D. They very often ignore architectural characteristics such as number of doors, openings, and windows. The granularity of the models is still very low, i.e., they do not take into consideration moveable obstacles (such as furniture), of functional spaces such as “coffee corner,” “resection area,” etc. Still most of the topological models used for navigation are predefined, are pre-computed, and cannot reflect dynamic changes as closing because of renovations. There is a vast amount of research in the area of indoor navigation and localization. Several conferences have been organized annually by various international organizations (ACM SIGSPATIAL, ISPRS, LBS, ICA, etc.). For example, the Indoo3D conference organized in December 2013 discussed topics related to indoor model definition, model generation, indoor localization, and indoor navigation applications.
Agreeing on standards for indoor models is one of the most investigated topics. It is well understood that standards will speed up the application development. Some researchers take into consideration not only the internal structure of a building but also the manner people can be localized indoors to be able to give directions. Commonly geographical coordinates do not make sense to humans. Humans, however, understand expressions such as “10 m left from the door” and “at front of the restaurant.” Xiong et al. (2013) presented the work on a multidimensional indoor location and information model, which aims to define absolute, relative, semantic, and metric expression of location. The model is complementary to 3D concepts such as CityGML and IndoorGML and is accepted as Chinese standards for coding location. Research on semantic expression of spatial relationships, directions, and locations such as “in room 321,” “on the second floor,” as well as “two meters from the second window” and “12 steps from the door,” has been discussed by a number of researches, e.g., Billen et al. (2014)
As mentioned previously, the 3D indoor models can be generated in various ways. Becker et al. (2013) presented an approach based on shape grammars applied to point clouds. Shape grammars have been proven to be successful and efficient to deliver volumetric LOD2 and LOD3 models, and the next challenge is its application to indoor modeling, i.e., LOD4 models. In building interiors, where the available observation data may be inaccurate, the shape grammars can be used to make the reconstruction process robust and verify the reconstructed geometries. The potential benefit of using the grammar as a support for indoor modeling was evaluated in the study based on an example in which the grammar has been applied to automatically generate an indoor model from erroneous and incomplete traces, gathered by foot-mounted MEMS/IMU positioning systems.
Point clouds are widely used for generation of 3D indoor models. They can be created using difference range techniques or from images and videos. Obtaining the vector model can be also done using many different approaches and algorithms. El Meouche et al. (2013) investigated automatic reconstruction of 3D building models from terrestrial laser scanned data. They proposed a surface reconstruction technique for buildings by processing data from a 3D laser scanner. Funk et al. (2013) presented a paper on implicit scene modeling from imprecise point clouds. The authors stated that when applying optical methods for automated 3D indoor modeling, the 3D reconstruction of objects and surfaces is very sensitive to both lighting conditions and the observed surface properties. This ultimately compromises the utility of the acquired 3D point clouds. The authors presented a reconstruction method which is based upon the observation that most objects contain only a small set of primitives. The approach combined sparse approximation techniques from the compressive sensing domain with surface rendering approaches from computer graphics. The amalgamation of these techniques allows a scene to be represented by a small set of geometric primitives as well as generating perceptually appealing results. The resulting surface models are defined as implicit functions and may be processed using conventional rendering algorithms, such as marching cubes, to deliver polygonal models of arbitrary resolution.
Wohlfeil et al. (2013) expressed the importance of using multi-scale sensor systems and photogrammetric approaches in 3D reconstruction. The authors discussed that 3D surface models with high resolution and high accuracy are of great importance in many applications, especially if these models are true to scale. As a promising alternative to active scanners (e.g., light section, structured light, laser scanners, etc.), the authors believe that new photogrammetric approaches are attracting more attention. They use modern structure from motion (SfM) techniques, using the camera as the main sensor. Their research combined the strengths of novel surface reconstruction techniques from the remote sensing sector with novel SfM technologies resulting in accurate 3D models of indoor and outdoor scenes. Starting with the image acquisition, all particular steps to a final 3D model were explained in their study.
The most prominent topic in indoor navigation is indoor localization. The indoor localization is in demand for a variety of applications within the built environment, and an overall solution based on a single technology has not been determined yet. This research is developed rather independently from the indoor modeling. The focus is on the technology that would allow localizing a person in a building, and therefore the indoor model is used mostly for visualization of the location. In the context of localization, 3D indoor models have been used for improving the localization accuracy (Girard et al., 2011; Liu et al., 2015). Many different localization technologies are investigated indoors as well (Fallah et al., 2013). Much attention is given to WLAN applications, which does not require a person to carry specialized devices. Two research papers presented at the workshop focused on the use of Wi-Fi technologies in indoor positioning. Verbree et al. (2013) investigated how Wi-Fi based indoor positioning can be used in museum environment to navigate three categories of users: visitors, employees and emergency services. They compared two different Wi-Fi-based localization techniques. The first one is based on Wi-Fi scanners, i.e., Libelium Meshlium Wi-Fi scanner. The second method was the traditional Wi-Fi fingerprinting. In a similar research, Chan et al. (2013) worked on improving Wi-Fi fingerprinting by applying a probabilistic approach, based on previously recorded Wi-Fi fingerprint database. In addition, the authors developed a 3D modeling module that allows for efficient reconstruction of outdoor building models to be integrated with indoor building models. The architecture consisted of a sensor module for receiving, distributing, and visualizing real-time sensor data and a web-based visualization module for users to explore the dynamic urban life in a virtual world.
Research on algorithms for indoor navigation is also very intensive with the aim to adapt them to the human perception and understanding. Particular indoors, well-known outdoor strategies as the shortest and the fastest path might be not relevant, while the safest, or less crowded, might be of relevance. Applications that support indoor navigation and way finding have become one of the booming industries in the last couple of years. In spite of this, the algorithmic support for indoor navigation has been left mostly untouched so far, and most applications mainly rely on adapting Dijkstra’s shortest path algorithm to an indoor network. In outdoor spaces, several alternative algorithms have been proposed by adding a more cognitive notion to the calculated paths and adhering to the natural way-finding behavior (e.g., simplest paths, least risk paths). The need for indoor cognitive algorithms is highlighted by a more challenging navigation and orientation requirements due to the specific indoor structure (e.g., fragmentation, less visibility, confined areas) (Vanclooster et al., 2013).
It stores all data needed in one database and processes most calculations on the web server which makes the mobile client very lightweight.
The network used for navigation is extracted semiautomatically and renewable.
The graphical user interface (GUI), which is based on a game engine, has high performance of visualizing 3D model on a mobile display.
Facility management is an area of research, which is increasingly gaining attention. Building owners are actively seeking for models that can give answer to questions as “how much paint do I need for the renovation of floor x,” “what is the area of the window frames that have to be pained,” and “how many square meters of carpet do I need for room y.” Facility managers need to have information about pipes and cables in case regular checks and/or failures. Local governments, institutions performing taxation, and so forth are also becoming interested in systems, which can easily compute net areas and volumes of apartments and offices. All these questions usually require information about vertical elements, internal structure of buildings, and even “invisible” information about pipe and cables integrated in walls and floors/ceilings. IFC and CityGML are very often compared and discussed, but still there is no agreement which model is more appropriate. For daily building and facility management, IFC appear to be too heavy and complex and numerous solutions are investigated considering CityGML.
Several 3D indoor models have been developed with the ultimate goal of finding an intermediate solution between IFC and CityGML. Hijazi et al. (2012) presents a model that integrates the building structure concepts of CityGML with the IFC concepts to provide simplified 3D model for maintenance of utility networks. The model is accessed by a simple application, which allows facility managers to explore and query their electricity and water facilities (Fig. 7).
Monitoring of Indoor Environments
Internet of Things (IoT) will be a key concept in monitoring of indoor environments. The IoT concentrates on making every physical and virtual “thing” a publisher of information. The IoT approach enables “things” to publish information once a state change occurs in them or in predetermined intervals. For instance, in a building that implements the IoT concepts, a door will publish information such as “I am closed now!” or a light bulb will indicate “I am on at the moment.” In addition, the “things” will become capable of taking actions based on messages coming from other “things” or humans. A building will be considered as a living entity, and applications will require information from the “things” (i.e., real and virtual) and the “models” (such as City GML/IndoorGML) in real time. In essence, applications such as Smart Buildings would require the fusion of information acquired from multiple resources, such as things, models, virtual objects, and real objects. The efficient monitoring of indoor environments will be directly proportional with the effectiveness in provision and fusion of real-time information related to indoors. By the utilization of ubiquitous monitoring of indoors, the information regarding building elements would be available 24/7 regardless of the situation (i.e., which can be emergency or nonemergency). Building and city dashboard applications would be the main consumers of this ubiquitous information. Combining semantic information coming from the indoor models with IoT data provides advantages in answering the emergency scene questions such as “would you provide the average CO2 level in the rooms which are not affected by the fire?” and “would you provide the number of doors which are open in the floors that are affected by the flood?” As another example, in a fire response operation, an emergency responder will acquire information from the sensors located in each floor regarding the spreading of the fire; in response, he can then invoke the web services to interact with IoT Nodes which will then invoke the actuators to close the doors in certain floors to prevent spreading of the fire to other floors. Furthermore, machine-to-machine (M2M) autonomous interaction is also possible, and a sensor can collect information regarding the emergency situation and interact with another IoT Node to perform a preventive action. As another sample, sensors in the building can interact with the actuators to close doors to prevent some parts of the building from being flooded by water; in fact, if there would be people in these parts of the building, they can be trapped as they cannot get out. In this situation, the people in the rooms can interact with the IoT nodes (to control sensor and actuators) to let them out of that building part. IoT provides unique opportunities for indoor monitoring.
- Afyouni I, Ray C, Claramunt C (2012) Spatial models for context-aware indoor navigation systems: a survey. J Spat Inf Sci 4:85–123Google Scholar
- Azri S, Isikdag U, Abdul-Rahman A (2012) Automatic generation of 3D indoor models: current state of the art and new approaches. In: Proceedings of international workshop on geoinformation advances, MalaysiaGoogle Scholar
- Becker T, Nagel C, Kolbe TH (2008) A multilayered space-event model for navigation in indoor spaces. In: Lee J, Zlatanova S (eds) 3D geo-information science. Lecture notes in geoinformation and 3D Geo-information Sciences, Springer, pp. 60–77. Springer, Berlin/HeidelbergGoogle Scholar
- Billen R, Cutting-Decelle A-F, Marina O, de Almeida J-P, Caglioni M, Falquet G, Leduc T, Métral C, Moreau G, Perret J, Rabino G, San Jose R, Yatskiv I, Zlatanova S (2014) 3D city models and urban information: current issues and perspectives, European COST Action TU0801, EDP science, 130 pGoogle Scholar
- Fallah N, Apostolopoulos I, Bekris K, Folmer E (2013) Indoor human navigation systems: a survey. Interact Comput 25(1):21–33Google Scholar
- Liu L, Xu W, Penard W, Zlatanova S (2015) Leveraging spatial model to improve indoor tracking. In: Fuse T, Nakagava M (eds) ISPRS Arch Photogramm Remote Sens Spatial Inf Sci, XL-4/W5, pp 75–80Google Scholar
- Worboys M (2011) Modelling indoor space. In: Proceedings of the third ACM SIGSPATIAL international workshop on indoor spatial awareness, Chicago, pp 1–6Google Scholar
- Xiong Q, Zhu Q, Zlatanova S, Huang L, Zhou Y, Du Z (2013) Multi dimensional indoor location information model. Int Arch Photogramm Remote Sens Spat Inf Sci XL-4/W4:11–13Google Scholar
- Zlatanova S, Sithole G, Nakagawa M, Zhud’Q (2013) Problems in indoor mapping and modelling. Int Arch Photogramm Remote Sens Spat Inf Sci XL-4/W4:63–68Google Scholar
- Zlatanova S, Liu L, Sithole G, Zhao J, Mortari F (2014) Space subdivision for indoor applications. GISt Report 66, 2014Google Scholar