**Taking Advantage of Collective Intelligence and BIM-Based Virtual Reality in Fire Safety Inspection for Commercial and Public Buildings**

#### **Daxin Zhang 1,**†**, Jinyue Zhang 2,\*,**†**, Haiming Xiong 3, Zhiming Cui <sup>4</sup> and Dan Lu <sup>4</sup>**


Received: 25 October 2019; Accepted: 21 November 2019; Published: 24 November 2019 -

**Abstract:** Commercial and public buildings are more vulnerable to fires because of their complex use functions, large number of centralized occupants, and the dynamic nature of the use of space. Due to the large number of these types of buildings and the limited availability of manpower, annual fire inspections cannot ensure the continuous compliance of fire codes. A crowdsourcing application, iInspect, is proposed in this paper to harvest collective intelligence in order to conduct mass inspection tasks. This approach is supported by building information modeling (BIM) based virtual reality (VR) and an indoor real-time localization system. Based on the International Fire Code and 27 fire inspection checklists compiled by various local authorities, a generic list of inspection items suitable for iInspect is proposed, along with a reputation-based monetary incentive model. A prototype of iInspect was created for Android mobile phones, and a case study was performed in an office building in Tianjin, China, for verification of this crowdsourcing inspection approach.

**Keywords:** fire safety inspection; building information modeling; real-time location system; smartphone; crowdsourcing

#### **1. Introduction**

As a major cause of injury and death across the world, fire is an important hazard that deserves special attention in risk management for buildings. A total of 237,000 fires were reported in 2018 in China, resulting in 1407 deaths, 798 injuries, and direct property losses of about 3.67 billion Chinese Yuan [1]. Some severe fires have devastating consequences. In Table 1, which lists the major fires in China over the last two decades [2], one can find that most of the severe fires that resulted in a large number of casualties occurred in commercial or public buildings.

The rapid and continuous economic development in China over the last four decades has led to large-scale urbanization. The intensive use of land in urban areas has greatly increased the prevalence of large-scale buildings, mainly complex commercial buildings such as shopping malls as well as public buildings such as hospitals. These types of buildings are deemed to have a higher risk of fire. First, the use of the interior spaces in these buildings is more complex. For example, a large commercial complex has multiple business types such as shops, restaurants, and cinemas. Second, due to the large scale and capacity of these buildings, the large number of occupants is concentrated in specific areas

and is not easy to evacuate effectively in the event of a fire. Third, the highly dynamic nature of these buildings makes fire prevention management more difficult. For example, it is not easy to track and manage the storage of flammable goods in shops.


**Table 1.** Major structural fires in China over the past two decades.

The design, construction, and operation of numerous building service systems are increasingly sophisticated, especially for large-scale commercial and public buildings. In order to reduce the number of fires, fire safety has been at the heart of legislation on building regulations [3]. Fire safety codes are often incorporated into local building codes and have been formulated by legislators following the advice of politicians and industry experts. There are also many safety-related requirements in current occupancy standards. Even if the building's systems strictly follow all regulations, however, one cannot assume the systems will run forever without creating any safety hazards. Dieken [4] states that in fire situations, one third of safety barriers are not in proper working order because of improper inspection or inadequate testing or maintenance. As such, building inspection is an important process for ensuring compliance with all safety regulations in everyday operation.

A building fire inspection is usually performed by a fire inspector, who makes a professional judgment about whether a building meets the requirements in a building fire code. Best inspection practices aim at identifying and correcting problems before they can have a negative impact on the building or the general public. A fire inspection program has significant consequences for building owners and property managers as well as the insurance companies that assume some risk or liability for the building, even if the inspection process is not be appreciated or even noticed by the people who use the building.

Numerous research studies acknowledge the significance of fire safety inspection, and Table 2 lists some examples of studies in the literature. A popular topic of research related to fire safety inspection is the functionality of fire safety facilities. Sobral and Ferreira [5] proposed a qualitative approach of a fault tree analysis to assess the availability of a fire pump system, taking into account the specificity of dormant systems and the importance of the frequency of inspections or tests of these systems. Sobral and Ferreira [6] also stressed the importance of tests, inspections, and maintenance operations in the context of a fire sprinkler system. They proposed a methodology based on international standards and supported by test/inspection reports to adjust the frequency of these actions according to the level of degradation of the components and with regard to safety. In 2004, Sierra et al. [7] conducted a field study in Spain, visiting 164 hotels and analyzing the status of their fire protection systems.

Along with advances in information and communication technologies, better approaches are adopted for fire safety inspection simulation and education. Han et al. [8] proposed a simulation training system for fire safety inspection to be used in colleges and fire brigades, based on three key technologies: three-dimensional (3D) geographic information systems, 3D modeling technology, and visualizations of human-computer interaction. Lee et al. [9] investigated the effectiveness of using an on-line training program for improving awareness of fire prevention in hospitals and found that visual-based on-line training can greatly improve healthcare workers' knowledge about fire prevention and evacuation. Zhang et al. [10] proposed an architecture for a smart system that provides customized instructions for an occupant to evacuate a building in the event of a fire; the system incorporates 3D building models with the support of 3D building models and a real-time location system (RTLS). Some researchers have focused on inspection criteria. Lo and a co-worker discussed how a building safety inspection system should accommodate buildings constructed in accordance with previous prescriptive standards by a fire risk assessment to determine the safety index [11] and from the index to rank the priority of improvement actions [12]. Wen and Fan [13] analyzed the catastrophic behavior of a flashover and developed a more appropriate way to determine the inspection criteria for flashover in compartment fires.

Associated with the criteria of fire safety inspection is a fire safety assessment. For large-scale commercial buildings, Liu et al. [14] proposed a fire risk assessment system based on the fire characteristics and the state of maintenance of fire equipment in the buildings. The proposed assessment system focused on evaluating the safety performance of the fire protection systems in the buildings. The Central Business District of Dares Salaam City carried out a study to assess urban fire risk with respect to public awareness on the use of firefighting facilities and preparedness in the event of the outbreak of fire [15]. The policy and making of laws and regulations with regard to fire safety inspection is also an active topic of research. In collaboration with the Atlanta Fire Rescue Department, Madaio et al. [16] developed the Firebird framework to help municipal fire departments identify and prioritize fire inspections for commercial properties by using machine learning, geocoding, and information visualization. Shao et al. [17] conducted inductive analysis and a survey questionnaire to discuss the integration of inspection and reporting systems for fire safety equipment and public building security. This study attempted to clarify the ambiguity between firefighting and building safety and to provide initial results to support amendments to related laws and regulations.



Existing research related to fire inspection does not consider continuous compliance to fire codes. Fire inspections are usually scheduled periodically (for example, on an annual basis). On one hand, good inspection programs require a significant amount of investment in human resources to conduct the inspection, and the inspectors are presumed to know the myriad code requirements, procedures, and processes. In reality, fire inspection programs may be inadequately staffed, may not be supported with a strong enough level of commitment by the local authority, or simply may not exist. On the other hand, some items require inspection at more frequent intervals because of the dynamic features of their status. For example, exit doors could be improperly locked during the daily operation for security

purposes and are only unlocked temporarily during inspections. Another example is the storage of highly hazard contents that may be present during the normal daily operation of a facility.

To focus on the inspection of items in commercial and public buildings that have dynamic features, this study proposes a social approach that combines collective knowledge from the general public with building information modeling (BIM)-based virtual reality (VR) technology to identify possible issues related to fire safety. Based on this approach, a prototype mobile app called iInspect was developed. With the help of an indoor real-time location system (RTLS), iInspect notifies occupants when they are near an item in the building that is fire safety—related and could be inspected (for example, a fire door). If a person has some spare time and would like to perform an inspection, the app will display the VR scene on the user's mobile phone to help him/her locate the items to be inspected and will provide a checklist and some additional information such as images of the correct status of the item and common problems or issues with the item. App users can perform an inspection and upload a report along with the necessary photos for verification purposes. The office of the appropriate authority will be notified and can take the necessary action(s) according to the severity of the issue.

In the following section of the paper, we will review related concepts including crowdsourcing and collective intelligence, BIM and VR, and RTLS. The system design for iInspect will be presented in terms of framework design, inspection content, and incentive mechanisms for social participation. A case study is conducted to examine and verify the usability of this approach, and the limitations of the system and possible uses of this approach in other inspection-related areas will be discussed.

#### **2. Related Work**

#### *2.1. Crowdsourcing and Collective Intelligence*

Crowdsourcing is a portmanteau to describe how businesses use the Internet to outsource work to the crowd. After studying more than 40 definitions of crowdsourcing in the scientific and popular literature, Estellés-Arolas and González-Ladrón-de-Guevara [18] summarized crowdsourcing as a type of participative online activity in which a party proposes to the public via a flexible open call the voluntary undertaking of a task. Crowdsourcing in the modern conception is an information technology (IT)–mediated phenomenon, meaning that a form of IT is always used to create and access crowds of people [19]. Members of the public submit their work, money, knowledge, and/or experience to the initiating party and are compensated with monetary prizes or with recognition of their contribution. Usually the crowdsourced result is given back to the public as a kind of response, especially when there no monetary compensation is involved [20].

By properly using a crowdsourcing method, one can manage large jobs with thousands of workers or do small jobs that require just a single person. According to the problem being solved, crowdsourcing can be classified into following four types [21]:


social networks, especially for startup companies or innovative ideas that need money to realize them. Contributors can receive goods or services or they can give money as a donation.

The crowdsourcing work involved in this research is a type of microtask. In order to compensate for the lack of an adequate fire inspection workforce, the key point in this study is to hire the crowd to inspect many commercial and public buildings against fire safety-related requirements. In this case, the huge job of inspecting many buildings at more frequent intervals is divided into many small tasks of inspecting only one or two items in a given building. This makes a seemingly impossible goal more achievable. To accomplish the crowdsourcing, there is a need to create a platform that can engage interested people to participate in the fire inspection process. Contributors can be compensated by cash (if the initiating party has a budget) or by acknowledgement (or any other in-kind offering).

Collective intelligence (CI), known as the wisdom of the crowd, refers to the collective opinion of a group of individuals rather than that of a single expert [22]. Crowdsourcing usually generates CI, which provides better results than can be achieved by hiring one or a few experts. A typical example is Wikipedia, an online, collaboratively created, and maintained encyclopedia—for this type of project, it is impossible to achieve the desired quality in a traditional editing paradigm [23]. CI is not new to the Information Age, but the rapid expansion of mobile Internet has largely facilitated the development of social information sites that rely on collective human knowledge, such as Yahoo! Answers, Quora, Stack Exchange, and similar sites. A crowdsourcing mobile inspection app, such as the iInspect app proposed in this study, can better take advantage of collective knowledge. For example, a mistakenly reported violation could be flagged as incorrect by other users in order to make the inspection results more accurate.

#### *2.2. BIM Modeling and Visualization Using VR*

BIM is a "modeling technology and associated set of processes to produce, communicate, and analyze building models" used for the design, construction, and routine operation of buildings [24]. BIM has been used to improve product delivery, and systems based on BIM have been adopted to perform tasks such as clash detection in the design and construction industry [25], visualize and optimize schedules in four dimensions (4-D) [26], and to prepare quantity take-offs during the pre-construction phase [27]. As a BIM model can encapsulate large amounts of data and information accumulated in the design and construction phases, it is a good starting point for facility management. Many concurrent facility management systems such as ARCHIBUS use BIM to support the visualization of data and information in 3D spaces, for example, allowing users to click on a pressure gage in the model to access information such as the brand/model, installation date, purchase price, warrantee information, maintenance schedule, and other information. In this research, any information that is relevant to fire safety inspection is incorporated into the BIM model of the building and is associated with each piece of fire safety equipment.

For crowdsourced inspections, it is necessary to take BIM models of commercial and public buildings to a cloud-based service that can enable visualization on mobile phones. The online visualization of BIM models is different from that employed in models running on local workstations used in the design and construction phases. An online BIM model must be significantly more lightweight (in terms of memory footprint and central processing unit usage) than a locally stored BIM model that may be more than 300 Mb in size. Until Internet service on 5G mobile phones is widely available, lightweight BIM data will be required to the model for mobile phones in order to improve the user experience. Moreover, all mobile phones do not use the same operating system (OS) and an online BIM model must be compatible with various OSs. For example, smartphones used by people in a building may use Apple iOS or Google Android, while the workstations in the fire safety management office might be running on a Microsoft Windows or Apple macOS operating system. It is imperative for the transfer of information on these platforms to be seamless.

The system developed in this study uses the Forge Viewer platform to enable BIM visualization on the cloud. This platform was formerly known as Large Model View, since it allowed the handling of 3D models that were too large to run on Autodesk Design Review. Forge Viewer converts information from a simple vector format (SVF) to Web Graphic Library (WebGL) format for display on an Internet browser; this allows people with mobile devices that use any OS to view information without having to install a plug-in. WebGL is an application programming interface (API) that allows users to create 3D graphics in their device's web browser, regardless of the software package used and without having to pay royalty fees [28], and some researchers have used this API to visualize BIM models online. For example, Xu et al. [29] developed a methodology for creating BIM model visualizations in three dimensions that combines the Industry Foundation Classes (IFC) data model using WebGL technology. In addition, Chen et al. [30] developed a cloud-based system architecture that enables very large BIM models to be viewed, stored, and analyzed for purposes such as facility management.

In order to help the building occupants to easily spot fire safety-related facilities, iInspect incorporates VR scenes along with the BIM model. Typically, VR is a simulated experience that allows users to become fully immersed in a virtual environment that could be similar to or completely different from the real world. Traditionally, a specially designed room with multiple large screens was required to create a VR environment. Currently, standard VR systems use head-mounted devices such as HTC Corporation's VIVE system to simulate a user's presence in a virtual environment [31]. In contrast to immersive VR, which is not practical to use for an app, the VR used in the proposed iInspect system is VR photography, which is the interactive viewing of a 360 ◦C circle or a photograph. Instead of using image stitching technology, hardware improvement has made the capture of VR photos much easier by using 360 ◦C cameras having two or more lenses. An Insta360 two-lens camera was used in this research to generate VR photos at places where some fire safety–related facilities are located that could potentially be inspected by iInspect users. Figure 1 shows the BIM model view and VR view of the perspective of a corridor in a commercial building.

**Figure 1.** (**a**) Building information modeling (BIM) model view of a corridor. (**b**) Virtual reality (VR) photograph of the same corridor from the same perspective.

#### *2.3. Indoor Real-Time Location System*

To locate iInspect users in real time, a real-time location system (RTLS) was used. iInspect will notify the user if any fire safety-related facilities are in nearby locations, and the user can view the information on his/her mobile phone and decide whether or not to perform an inspection. Accurate indoor RTLS is needed to inspire the crowd to effectively inspect fire safety-related facilities, as people will be more apt to take on an inspection job if they are notified by the system when some facilities are nearby, instead of having to search for nearby facilities on their own. There are several technologies currently used in indoor RTLSs, and these technologies are discussed in greater detail in the following paragraphs.

Radio frequency identification (RFID) has been adopted in many RTLS systems because it uses an inexpensive and flexible approach to identify objects and people. RFID is commonly used in applications for indoor settings, as its range of accuracy is 1–3 m; the global positioning system (GPS) has a lower range of accuracy (5–10 m), and it is more often applied in outdoor settings. For fire evacuation management, RFID would require all occupants of a building to wear RFID tags. It might be possible for an office building in which most building occupants are regular occupants; however, the majority of occupants of a shopping mall or a hospital visit only on an occasional basis.

Ultra-wide band (UWB) technology has an accuracy that ranges from 0.1 to 0.3 m, giving it greater accuracy than RFID. In addition, its short response time makes it applicable to settings that are indoor or out-of-doors. The main drawback is the high cost for implementing a UWB solution, which can make its use unfeasible for fire evacuation management [32].

While vision analysis has advantages over RFID and UWB because no tracking devices need to be affixed to the people or objects being tracked, it does have some drawbacks. Vision analysis needs access to a voluminous amount of labeled data sets to use for training prior to implementation. In addition, efficient visual analysis is difficult to achieve in an environment that is dark or dusty, which makes it problematic for evacuations during a fire, when smoke and flames may drastically reduce the accuracy of a vision-based location system.

Other systems—such as Bluetooth low energy (BLE), wireless local area networks (WLAN), Wi-Fi, ultrasound, lasers, radar, infrared, and magnetic marker fields—can be used for indoor localization, and each technology has benefits and drawbacks. In previous research on RTLS technology [10], the authors compared various systems reported in the literature (Figure 2) in terms of the range of accuracy, the affordability of the technology, and how easy the technology would be to implement. In Figure 2, the size and position of each are relative, as they are not based on precise values.

**Figure 2.** Comparison of major real-time location system (RTLS) technologies (where the size of each circle reflects how widely the technology has been adopted, the vertical axis indicates the range of accuracy, and the horizontal axis indicates the ease of implementation).

In this study, the proposed RTLS uses indoor localization that is based on BLE technology, which enables contextual information to be conveyed between various connected devices (such as between BLE sensors, smartphones, and computers) using a minimal amount of equipment (with regard to the size and cost of the required infrastructure) [33]. As a BLE-based system has low power requirements to operate, the tracking devices will have a long lifespan. In addition, since the BLE devices used in this system are small, they are considered to be a wearable device (it can be attached to the hardhats or safety vests of construction workers, for example). Despite its potential to be used in a wide range of areas, researchers and professionals have not considered BLE-based RTLS for very many application scenarios.

For buildings, such as the floor of the building shown in Figure 3, Bluetooth sensors can be placed in multiple locations on each floor, with a layout that depends on the room layout for each floor, as a single Bluetooth sensor will generally cover an area with a radius of 5–10 m. The Bluetooth sensors are the size of a coin and cost about US\$12 each, can be used for 6–12 months before the batteries need to be replaced, and are small enough that they will not have an obvious or unappealing appearance. Batteries should be checked at six-month intervals to ensure that they are working properly. A mobile phone can be used to establish a communication link between a Bluetooth-enabled device and the surrounding sensors, and the location of the smartphone can be determined using the received signal strength (RSS) method.

**Figure 3.** Bluetooth low energy (BLE)-based indoor RTLS.

#### **3. System Design for iInspect**

#### *3.1. iInspect Framework*

Figure 4 illustrates the framework of iInspect, where the solid lines represent physical movement and the dashed lines show the circulation of information in cyber space. iInspect will determine which building a user is currently occupying with the assistance of the GPS module in the user's mobile phone and the geographic information system (GIS) coordinates for the building. The iInspect user is able to see the 3D spaces of the building through the cloud-based lightweight BIM model, which should be updated in a timely manner following any changes that are made to the physical building. In a sense, the BIM model is the digital twin of the physical building. While the user walks through the building, iInspect will combine the real-time location of the user and the location of fire safety-related equipment as recorded in the BIM model, and the app will notify the user if there are any facilities nearby that can be inspected.

After being notified by iInspect, a user can browse information related to the fire safety-related facilities that are associated with 3D digital components in the BIM model. There are three types of information that would be available via the app. First is basic information, which includes facility attributes (such as the type and expiration date for a fire extinguisher) and the items that can be inspected along with inspection guidelines (i.e., how to determine the operational and defective status). Second, some VR photos of the real environment will be made available via the app. Using these VR photos, iInspect users can easily find facilities to be inspected and understand the correct status and common problems. Third, if any inspection reports have been generated by other users for the facility, all iInspect users can see those reports and are able to verify the content by performing an additional inspection. All of the information would be updated periodically to reflect the latest status of each facility. Based on this information, an iInspect user could perform an inspection, upload some photographs as a kind of proof, and file an inspection report. After the inspection, a new report or an updated report will be generated and, if a defective status is noted in the report, fire safety management officers will be notified by flags of various colors (which can be used to indicate, for example, different levels of severity).

Fire safety management officers will review all reports having flags and can give feedback to the users who filed the reports regarding the validity of those flags. iInspect users will receive rewards of different kinds for valid flags they have reported. At the same time, the building manager should be notified (and issued a ticket) if there are indeed some fire safety violations. If it is deemed necessary, fire safety management officers will make a trip to the building to perform an onsite inspection.

**Figure 4.** System illustration of iInspect.

#### *3.2. Inspection Content*

It is necessary to clearly define the inspection content for iInspect because not all fire safety related facilities in a building are accessible and able to be inspected by the general public (for example, to check if the sprinkler heads are free of grease, dust, rust, etc.). Based on the 2018 International Fire Code [34] and Fire Inspector's Guide [35], along with a systematic review of 27 fire safety inspection checklists made by different local fire departments, authors identified items suitable for iInspect; these items are summarized in Table 3.

It is noted that this list is only a subset of the fire inspection list that is used for regular annual inspection; thus, this crowdsourcing-based inspection can in no way replace a regular fire inspection. The focus in the list for iInspect are items that are publicly accessible and have a dynamic status of validation. When iInspect is applied to a specific setting, the list could be amended by local authorities to make it suitable for a particular circumstance.




#### **Table 3.** *Cont.*

#### *3.3. Incentive Model*

People's participation and willingness to contribute are critical issues that one must take into account when developing crowdsourcing solutions. Hence, people should be given some incentives to become part of this collaborative process [36]. Incentives can be intrinsic (personal enthusiasm or altruism) or extrinsic (monetary or in kind reward) [37]. Mohammad et al. [37] argued that intrinsic incentives are more positive than extrinsic ones in terms of the quality of the crowdsourcing project. Monetary incentives can speed up the attraction of participants, however, the payment assigning method can better affect the outcome's quality than simply increasing the amount of money [38]. In all cases, insufficient incentives result in dropping out from the crowdsourcing efforts [39].

There are many cases that people on the Internet take different task for intrinsic and extrinsic incentive other than money. This is more used in knowledge sharing website (for example, Wikipedia and wikiHow) and question answering websites (for example, Quora and Stack Overflow). Different non-monetary incentive types include:


A carefully designed incentive system is vitally important for the success of a crowdsourcing project. A reputation-based monetary incentive model is proposed for the iInspect app developed in this research. More money could increase participation but not necessarily the quality of the results and offering less money may lead to an insufficient number of participants accepting the task. As such, participants need to be paid based on their reputation score (RS), which is earned through the quality of their work. Some researchers have designed monetary incentive models, which integrate participant history/reputation for determining task allocation and payment of rewards [43,44].

In the incentive model for iInspect, the crowdsourcing project requester, i.e., the local fire safety management office, has to evaluate all inspection reports and return a rating called a feedback rating (*FR*) where *FR* ∈ {−1, 1} to participants. A rating of "−1" represents an unsatisfactory inspection report, where the inspection result contains mistakes or false fire safety violations. In contrast, a rating of "1" represents a satisfactory report, which means that violations were found to be valid. Every participant in iInspect is assigned an initial reputation score (*RS*) of 10; the *RS* is dynamically adjusted by the *FR* of a completed inspection task. The monetary reward (*M*) paid to participants is calculated as the product of a base payment (*BS*, which is US\$1 in this study) and *RS*, i.e., *M* = *BS* ∗ *RS* | *RS* >0, *FR* = 1. This means that a participant will be paid more if he/she has a higher reputation score. There are two conditions required to obtain a positive value for *M*. The first one is that the *RS* must be a positive value. If a participant has a negative value for *RS*, it means he/she submitted too many unsatisfactory inspection reports; in order to be paid, he/she must perform more inspections and submit satisfactory inspection reports to bring the reputation score back to a positive level. A participant will also be paid for inspection work only when a report is determined to be satisfactory (i.e., *FR* = 1). In short, after each inspection, the reputation score for a participant will be updated, and reward money will be calculated and paid based on the feedback rating.

Aside from the monetary incentives calculated using reputation scores, some non-monetary incentives are also embedded in this crowdsourcing application. First, by participating in fire safety inspection tasks, a user can develop their personal awareness and fire safety skills. This is a very important personal development and is valuable for personal safety. Second, it builds up high regard and honor in a user's social network. In this research, iInspect links a user's account with his/her account in a major social media platform in China, WeChat, which allows people to know how well their friends are performing in iInspect. A user will be very proud of completing a number of inspection tasks and having a good reputation in the system, because this means he/she contributes significantly to public safety.

#### **4. Case Study**

#### *4.1. BIM Model and VR Scene Dvelopment*

The case study conducted by the authors used an office building located in Tianjin, China, for verification of the social inspection approached proposed in this research. The floor plan of the 34th floor of the selected building is shown in Figure 5a. A BIM model was first built by importing the construction model for the building into Autodesk Revit and removing any irrelevant information on the facilities related to fire safety that were located on the 34th floor. Next, the BIM model was exported to Autodesk Forge, where a lightweight 3D model for the 34th floor of the building was rendered (as presented in Figure 5b).

**Figure 5.** Commercial building that was used in the case study: (**a**) Floorplan of the 34th floor of the building. (**b**) BIM model of the 34th floor as viewed in Forge (with navigation tools).

VR photos for the building, which were obtained using a dual-lens Insta360 VR camera, were collected at the following locations: two exit doors, two fire doors, floor map sign, four emergency lights, six exit signs, one fire alarm call point, two fire extinguisher boxes, sixteen sprinkler heads, four smoke detectors, and four storage locations. VR photos allow iInspect users to view the environment in a 720 ◦C perspective (360 ◦C horizontally and 360 ◦C vertically). The VR photos are very helpful, as they assist the participants in locating the facilities to be inspected by enabling them to compare the physical environment to the VR environment. Figure 6 shows the VR images of a fire door and a fire extinguisher box.

**Figure 6.** (**a**) VR image of a fire door. (**b**) VR image of a fire extinguisher box.

#### *4.2. Implementation of BLE-Based RTLS*

Due to budget restrictions, only a single floor of the commercial building was used in the case study. Bluetooth sensors (Figure 7a) were used to implement the RTLS for the RTLS. The installation included 26 Bluetooth sensors in the 34th floor of the building (the locations of the sensors can be seen in Figure 7b), and the coverage area for each sensor had a radius of 5–10 m. A cloud server was used to determine the location of a smartphone user in the building based on the data exchanged between the smartphone and the Bluetooth sensors. The Bluetooth system considered in the case study was accurate to within a range of 1–5 m. The actual range in other settings depends on the quantity of sensors deployed and the locations of these sensors, as well as on the obstructions in the building (such as walls) and the material composition of these obstructions, the sensor power, and other factors. After this testing system was deployed, it was calibrated by comparing the coordinates for a building occupant that were calculated by the system to the actual coordinates of the occupant in the building. Since the building had a structure made of reinforced concrete, the partition walls on this floor of the building were not considered to be a serious obstruction. As tested, this system was able to achieve an accuracy of between 1.5 and 1.8 m.

**Figure 7.** (**a**) Bluetooth sensor employed in this study and (**b**) layout of 26 Bluetooth sensors.

#### *4.3. Typical Reporting Process*

The prototype iInspect app (Figure 8) developed in this study for smartphones using Android did not include functions (such as account management or other functions) that were not directly related to the verification The researchers presumed that an iInspect user had previously downloaded and installed the prototype and had set up a user account. In addition, iInspect is able to use GPS information from the smartphone to determine if a user has entered a building where inspection information for the building has not previously been downloaded using the app—in such a case, the app will be able to prompt the smartphone user to download inspection information for the building from the external server to view any possible inspection tasks.

**Figure 8.** Prototype application of iInspect: (**a**) Possible inspection locations. (**b**) Indoor navigation to the inspection location. (**c**) Items to be inspected. (**d**) Account overview. (**e**) Inspection details. (**f**) Rank details.

When an iInspect user is near a fire safety-related facility, the system will push a notification to the user, and an overview of checkable facilities will be displayed on the phone that contains additional information about the facilities to be checked. As shown in Figure 8a, each bubble in the floor plan represents a certain type of facility available for inspection; when a bubble is selected, the basic information about the associated facility and the amount of reward money for the inspection task will displayed at the bottom of the screen.

After clicking the "Go" button, iInspect will show an interface of indoor navigation to guide the user to locate the facility, as shown in Figure 8b. The "VR" and "BIM" buttons in the navigation view allow the user to switch between the BIM view and the VR view if a VR view is available.

If the "Inspect" button is clicked, a checklist will be displayed, as shown in Figure 8c. By clicking the "i" icon in front of each item, a detailed explanation and the associated illustrations will be displayed at the bottom of the screen. If the user believes there is a violation for a given item and clicks the "cross" icon, the system will require a photograph to be uploaded as proof of the violation. If all items pass the inspection, the system will prompt the user to upload a photo of the entire facility along with a timestamp as proof of the inspection.

A participant can upload an inspection report to the server by clicking the "Done" button, and fire safety management officers will be immediately notified if any safety violation is reported. As mentioned previously, an inspection report will be rated, and the RS and monetary reward (*M*) will be calculated. An iInspect user can view his/her account to check the value of the RS and the balance of *M*, as well as the history of inspections, as shown in Figure 8d. Inspection details will be displayed (Figure 8e) after the participant clicks any inspection history record. In addition, the ranking of the user with respect to his/her friends (Figure 8f) will also be displayed, and this information is shared on social media.

#### **5. Conclusions**

Fire safety is a vitally important issue for all buildings. Annual fire safety inspections performed by the local fire safety management office is the current practice to ensure there are no violations for any fire safety-related requirements. Commercial and public buildings are deemed to be more vulnerable to fires because of their complex use functions, large number of centralized building occupants, and the dynamic nature of the use of these buildings. Ideally, there needs to be an approach to track the status of all fire safety-related facilities in a real-time manner, or at the very least to conduct safety inspections on a more frequent basis. However, the reality is that many cities lack a sufficient number of skilled fire safety inspectors to ensure continuous compliance of all fire codes for the large number of commercial and public buildings under their jurisdictions.

This paper proposed a crowdsourcing application, iInspect, which can be used to recruit members of the general public to carry out fire safety inspection tasks through the assistance of BIM + VR and indoor RTLS. iInspect assigns inspection tasks to its users when an indoor RTLS detects there are fire safety-related facilities nearby, and it employs BIM + VR technology to help the user find the facility and understand what needs to be inspected. In this way, iInspect can harvest collective intelligence, which is the key benefit of a crowdsourcing project.

Since the general public cannot access and inspect all fire safety-related facilities, this research summarized a list of items that can be easily checked by a typical building occupant. The items on the list are all accessible by the general public and have highly dynamic features, including building access and outdoor premises, exits and escape routes, electrical systems, emergency lighting, fire extinguishers, fire alarm systems, fire sprinkler systems, heat-producing devices and appliances, maintenance of building areas, and smoke and carbon monoxide (CO) detectors.

An appropriate incentive model is the key for the success of any crowdsourcing project. This research proposed a reputation-based monetary incentive model for iInspect. An iInspect user will gain a higher RS when he/she takes on more inspection tasks and produces high-quality inspection reports. The actual money an iInspect user will earn is the product of a base amount and his/her RS, which means that a user with a higher reputation score will earn more money than other users when performing the same inspection task. Some other non-monetary incentives are also considered in this research, such as the development of personal skills and altruism/honor.

A prototype application of iInspect was created for mobile devices using the Android platform, and a case study was conducted to verify the developed system by considering an office building located in Tianjin, China. A BIM model was developed along with VR photos of all major items to be inspected. An indoor RTLS was deployed using BLE technology, and an accuracy of 1.5–1.8 m was achieved. The prototype app, which runs on a mobile phone, follows the iInspect framework shown in Figure 4 and has verified major functions.

This crowdsourcing approach can be applied to many other inspection circumstances with a lack of inspectors that present a highly dynamic status of compliance. For example, the on-site safe management of a construction project is a challenge due to its dynamic nature. By using a crowdsourcing approach, every on-site worker with a smartphone has the potential to be an inspector, and any potential risk could be identified and reported in a timely manner. Similarly, for some infrastructure facilities such as highways, the spatial distribution of networks makes continuous inspection/assessment of the network condition difficult and costly. This issue will be well handled if an adequate number of infrastructure users can be recruited as defect inspectors.

There are some limitations in this research. First, indoor real-time location technology is still in its early stages and there is much room for improvement in terms of accuracy and cost. As a result, an indoor RTLS cannot be readily implemented in most buildings; thus, push notifications of nearby checkable facilities and indoor navigation could be functions that may not be available in most buildings, and users will need to find facilities on their own using the BIM model, even a 2D map, as a reference. Second, the open nature of the crowdsourcing approach presents an opportunity for individuals who exhibit antisocial behavior to participate in the inspection process. For example, people may upload incorrect inspection reports deliberately. Therefore, the quality control of such a crowdsourcing inspection system is the next logical topic of study. Third, considering that non-expert general public many have limited ability to perform inspection tasks, local authorities may use a customized version of Table 3 in the early stage of implementation. Last but not least, the effectiveness of the incentive model employed in this research needs to be validated by implementing the iInspect app on a large scale. The prototype application was tested by only a few users, and it is not possible to examine the effectiveness of the incentive model when using such a small sample size.

**Author Contributions:** Conceptualization, J.Z. and D.Z.; funding acquisition, J.Z.; investigation, D.Z. and J.Z.; project administration, J.Z.; software, Z.C., D.L. and H.X.; writing—original draft, D.Z. and J.Z.

**Funding:** This work was supported by the National Natural Science Foundation of China under Grant 71272147 and Tianjin Science and Technology Committee under Grant 15ZXHLSF00070.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).
