A BIM-Centered Multi-Source Image Fusion Framework for Remote Client Site Visits

Abstract

Clients need to visit project sites periodically during construction to visualize progress and identify deviations from expectations. However, physical site visits are time-consuming, costly, and potentially unsafe, especially for remote and overseas projects. More fundamentally, existing remote-site-visit solutions focus primarily on automatic recognition and visualization, while insufficiently addressing the scientific challenge of how heterogeneous, dynamic site data can be fused and operationalized to support timely, collaborative decision making. This research proposes a framework for clients’ remote site visits. It develops an RASE system that enables multi-source data fusion and real-time collaborative decision support by integrating UAVs, 360° cameras, BIM, and VR/AR technologies. RASE allows clients to synchronize real-world visual data with BIM models within predefined scenes, annotate issues directly on BIM components, and seamlessly switch among heterogeneous image-capture sources to maintain situational awareness in highly dynamic construction environments. The proposed framework emphasizes an operational data-fusion mechanism and an interaction paradigm that reduces the cognitive and coordination burdens of remote decision making. A case study shows that RASE reduces site-visit time by 78.0%, though initial equipment costs increase total expenses by 44.1%. Sensitivity analyses indicate that projects with greater remoteness or higher visit frequency significantly improve both time and cost effectiveness. The core contribution of RASE lies in enabling a scalable, operational data-fusion mechanism that supports collaboration for remote site visits, with the associated issues for the corresponding BIM components. Automatic image and voice recognition functionality may be incorporated with RASE to improve the efficiency of system control, textual input, and BIM association in the future.

1. Introduction

During the construction phase, timely and accurate monitoring of onsite operations can provide immediate insight into project-specific issues. Systematic monitoring helps to quickly identify concerns and challenges on a project [1]. This process equips key stakeholders, such as clients, architects/engineers/contractors (A/E/C), with the necessary information to make prompt and effective project control decisions [2]. In construction projects, quality inspection objectives differ among the client, A/E, and contractor due to their distinct roles, with the client’s perspective primarily focused on project-level outcomes and value. From the client’s viewpoint, quality inspection aims to confirm that the completed work satisfies contractual requirements, functional performance, safety expectations, and long-term operational value, rather than to monitor detailed construction specifications and processes [3,4]. The A/E supports this objective by conducting conformance-oriented inspections to verify that construction complies with design intent, specifications, and applicable codes, acting as the client’s technical representative but not assuming responsibility for day-to-day quality control [3,5]. In contrast, the contractor’s quality inspections are process-oriented and operational, focusing on defect prevention, workmanship control, and compliance during execution to avoid rework, delays, and cost overruns [3,4]. Together, these differing objectives explain why client inspections emphasize decision making and acceptance at the project level, while relying on A/E oversight and contractor quality control for technical and operational assurance [6].

Currently, most site assessments [5,7] are conducted either by (1) physically traveling between the site and trailers to access paper-based drawings and specifications, which are then used to generate updated bases for evaluations, or (2) manually accessing and searching plans via tablets. This latter method requires the prior manual generation of specific 3D-rendered views for each inspection task. Given the thousands of elements and hundreds of activities on a construction site, this process can be highly time-consuming.

Numerous studies have attempted to automate the recognition of project progress at the activity, operation, or entity levels using various technologies. For instance, Turkan et al. [8] utilized point clouds to track secondary and temporary concrete construction objects, such as formwork and rebars. Bosché et al. [2] combined point clouds and BIM to monitor Mechanical, Electrical, and Plumbing (MEP) installations. Other research has focused on automatically recognizing entities on construction sites, including dump trucks, excavators, and workers [9,10]. In the analysis of operational productivity, Rezazadeh Azar et al. [11] estimated dirt-loading cycles by detecting and tracking excavators and dump trucks simultaneously. Yang et al. [12] monitored concrete placement activities by tracking the crane jib through 3D pose estimation, distinguishing between concrete placement and non-concrete material movement. Peddi et al. [13] tracked workers tying rebar using blob matching, extracting skeletons for pose estimation, and classifying their work status. Siebert and Teizer [14] used 3D mapping for surveying earthwork projects using an unmanned aerial vehicle and mobile 3D mapping.

In general, monitoring project-level information involves characterizing the extent to which construction plans are being followed and when performance deviations occur. Various manual or semi-automated methods have been proposed for progress monitoring using appearance data. Some research has attempted to monitor project progress or quality by comparing time-lapse images [15] or by using a 4D BIM [16,17,18,19]. In addition, others have focused on the automatic generation of as-built models, such as point-cloud models based on multi-view stereo cameras and voxel coloring algorithms [20] or improved image-based 3D reconstruction [21].

As highlighted by Yang et al. [6], despite substantial advancements over the past decade, many challenges remain unresolved. For instance, from a sensing perspective, most current efforts to generate point clouds still rely heavily on laser scanners. Although image- and video-based 3D reconstruction methods have shown promise, the accuracy and completeness of vision-based 3D reconstruction still require significant improvement. In construction entity detection, multiclass detection has not yet been well-developed, with studies focusing on only a limited set of entities, such as workers, excavators, dump trucks, loaders, and tower cranes.

Recognizing the insufficiencies and limitations of current research and development, this research focuses on developing the client’s visualization and control system to enable remote monitoring of construction sites. While site inspection still requires a physical presence on the site, current research and development do not yet make remote inspection feasible for practical use in project inspection at the activity or component levels. However, a remote site visit at the project level may be feasible and helpful for the client. It may also be useful to the project contractor’s executives who oversee multiple projects at remote locations. Site visits for a client are mainly to visualize project progress, identify problems and deviations, and communicate with people to ensure expectations are met or make necessary change orders. A remote-site-visit system should save time and cost because of the reduced frequency of traveling to the sites. It should also reduce the client’s safety concerns, as the client is not as familiar with the sites as the AEC staff.

According to current BIM-based best practices, as presented by McCarthy Construction [22], resolving a performance issue can take between 0.5 and 5 h. This timeframe includes tasks such as taking notes and photos, contacting relevant engineers, locating and analyzing project data, sending specifications and BIM screenshots, and instructing field teams to address the issue. Miscommunication can cause these steps to be repeated several times, and the estimated time does not account for travel between locations.

There are several system developments related to remote site visits on some other aspects, rather than the facilitation of client visits. Several researchers [23,24,25,26] have developed VR or Web systems to educate students or train staff on the construction process, safety measures, or reading plans. The virtual models in these works were not necessarily real time. Sun et al. [27] have even used four-legged robots to capture images from construction sites to help teach in classrooms. The application of trending technologies such as UAVs and AR/MR has also been investigated across different application areas within the AEC industry. De Melo et al. [28] developed a protocol for the data collection, processing, and analysis of visual assets (i.e., photos and video recordings) captured by UAVs over two surveyed job sites to improve safety inspections by providing better visualization of working conditions. Park et al. [29] proposed a framework to help manage construction defects using BIM and AR based on a pre-built ontology-based data collection template. Hou et al. [30] investigated how much-improved assembly productivity and performance can be achieved by lowering cognitive workload via AR (i.e., providing workers with 3D models instead of 2D drawings).

Research on construction site management increasingly positions Unity (Unity3D v2023.1.22f1) as a practical real-time engine that operationalizes BIM data for automated visualization, simulation, and safety management. Unity is commonly used to transform static BIM models into interactive construction environments that support automated rule checking, sequencing visualization, and risk communication [31,32,33]. By enabling real-time rendering, physics simulation, and user interaction, Unity-based systems align well with automation objectives by reducing manual interpretation of drawings and enhancing machine-supported decision making in construction planning and safety analysis.

A dominant research stream focuses on Unity-based VR safety training, which directly supports automated and standardized safety management processes. Empirical studies report that Unity-enabled VR environments improve hazard recognition, situational awareness, and training consistency when compared with conventional instruction methods [31,34,35]. These systems automate the delivery of repeatable, scenario-based safety training and enable systematic collection of performance data for analysis, consistent with the data-centric focus of construction visualization. However, studies also identify technical constraints, including BIM-to-Unity interoperability, model complexity, and the fidelity of simulated site dynamics, which limit scalability and automation efficiency [36].

Recent studies extend Unity applications toward automated and data-driven construction systems, including digital twins, real-time monitoring, and location-aware safety control. Frameworks that integrate Unity with BIM databases, sensors, and tracking technologies enable semi-automated updating of virtual construction sites to reflect real-site conditions [33]. Systematic reviews emphasize that while Unity-based VR and simulation tools demonstrate strong potential for automating safety training and visualization workflows, further research is needed on standardized data pipelines, multi-user collaboration, and longitudinal validation of safety outcomes to meet industry-scale automation requirements [37].

The most similar existing work to this research, in terms of technologies used and objectives, is probably Ammari and Hammad’s work [38], which proposed a framework to support remote collaboration and visual communication between field workers and office managers for maintenance inspections. The proposed framework integrated multisource facility information, BIM models, and feature-based tracking in an MR-based setting to retrieve information-based inspection schedules and the field worker’s location, visualize inspection and maintenance operations, and support remote collaboration between the field worker and the office manager. The main differences between Ammari and Hammad’s work and this research are as follows.

1.

This research focuses on the construction process instead of the maintenance process.

2.

The scope of the intended remote visit is to oversee construction at the project level, not limited to safety issues.

3.

The proposed framework can accommodate the BIM model rather than relying primarily on UAV imagery, project drawings, and specifications.

4.

The target user is the client, not the job manager. This research focuses on visual, oral, textual, and BIM-based communication to help the client remotely compare the physically constructed components on site with those in BIM and to communicate with on-site personnel, thereby reducing the potential risk of visiting the site in a dynamic environment.

5.

Most importantly, the proposed framework uses an innovative yet comparatively simple approach to enable switching among multiple image recording sources (e.g., UAV, 360° camera, smartphone), which are now commonly used on construction sites.

In summary, despite extensive research on automated progress monitoring, visual sensing, and BIM-based visualization, a critical gap remains in understanding how heterogeneous and asynchronous site data can be fused in real time to support collaborative decision making by non-technical stakeholders, such as project clients, under highly dynamic construction conditions. Most existing systems either emphasize automated recognition or provide visualization tools without addressing how users cognitively and operationally integrate multiple data streams during remote collaboration. As a result, current solutions remain limited in practical adoption for client-level project control.

This study addresses this gap by reframing remote site visits as a problem of multi-source data fusion and collaborative decision making, rather than a problem of sensing accuracy or visualization fidelity alone. The proposed RASE framework investigates how BIM, real-time visual streams, and human communication can be operationally integrated to support timely, shared understanding and decision making in remote, high-dynamic construction environments.

2. Research Problem and Methods

On-site visits for project clients are not only costly and risky but also represent a broader information management challenge: clients must integrate fragmented visual observations, verbal explanations, and BIM information to form actionable and traceable decisions under time constraints. Existing remote-visit approaches typically rely on post-processed image repositories or single-stream live video, which limits situational awareness and hinders real-time collaboration.

From a scientific perspective, the key problem is not the lack of sensing devices, but the absence of an effective mechanism for fusing heterogeneous data sources—UAV imagery, ground-level video, BIM representations, and human annotations—into a coherent decision-support environment. Moreover, remote stakeholders must be able to interact with this fused information collaboratively, without imposing excessive operational or cognitive burden.

Current developments to enable the client to visit sites remotely use advanced devices such as UAVs and multiple-view cameras. However, they still rely on a repository of large image data for post-processing at the job site office or headquarters. Accessing large image datasets is inconvenient for the client, and real-time vocal and visual communication of identified problem areas to on-site personnel is difficult with this approach. The client visits the site mainly to visualize the project’s progress and ensure the expected need is met, rather than to verify that the construction meets the detailed design specifications, which are usually verified through inspections conducted by AE professionals. When the client finds a problem, they usually prefer to describe it orally to on-site personnel rather than communicate it in writing, drawings, or specifications. Thus, this research proposes a framework to support remote site visits for the project client and enable real-time audio-visual communication with on-site personnel by integrating data flows among the UAV, AR goggles, and the BIM model.

Based on the surveyed companies, there are two common approaches to how on-site engineers help clients virtually visit construction sites. First, site personnel used UAVs or video cameras to record the project’s overall progress. The images and videos were categorized and stored at the job site office or on the cloud. Key images and explanations may be annotated with the BIM model at the job site office. The client then watched the presented materials and recorded the instructions or the problems for further clarification or discussion. The messages were then sent to the job site office. This approach has the advantage of allowing the client to compare site images with the BIM model. However, the approach requires extensive preparation and background work. In addition, the client cannot view the site or communicate in real time, and they cannot freely control viewing locations.

The other approach uses mobile devices to help clients virtually visit construction sites. Site personnel point-and-shoot the construction site on phones or tablets in real time, either during a teleconference with the client or via remote collaboration software. Engineers share the corresponding components of the BIM model as the recorded video stream goes on. The client may direct site personnel to change the shooting places via mobile phone. The client may also provide instructions or discuss the identified problems in real time. This approach has the advantage of not requiring a large data repository or inconvenient access to large image datasets. However, the matching between real-time images and the BIM model needs someone to operate for the client. Using mobile devices alone also limits the project’s holistic view that UAVs can provide.

This research develops a system named RASE (Remote Augmented Site Explorer) that leverages both of the aforementioned approaches. It allows clients to visit sites remotely via UAVs or multiple-view cameras, visually map images to corresponding BIM components, and engage in real-time communication, with the ability to annotate BIM components. As shown in Figure 1, on the right is the client’s office, where the client can remotely visit a construction site (on the left) using RASE, RASE comprises four main modules: image capture, remote collaboration, BIM/VR Integration, and AR modules.

Figure 1. The conceptual framework of RASE (Orange arrow: main data flow direction; Dashed square: remote site).

Conceptually, RASE can be viewed as a lightweight, interaction-centered data-fusion framework rather than a monolithic automation system. Instead of attempting to automate progress or defect detection, RASE focuses on enabling users to dynamically align, compare, and annotate heterogeneous data streams to support collaborative decision making. This design choice reflects the reality of high-dynamic construction environments, where incomplete, uncertain, and evolving information often makes full automation impractical.

Since the implementation and usage are better explained with a case, the following presentation begins with a survey case and then proceeds through each main module. This study used a local 12-story building project as the test case. The BIM model of the building was available from the owner in Autodesk’s Revit, and we helped add additional MEP (Mechanical/Electrical/Plumbing) information for demonstration purposes. The Revit BIM model of the case is shown in Figure 2.

Figure 2. BIM model of the studied case.

Figure 3 illustrates the prototype implementation of the RASE system, with an emphasis on the data and command flows among its four main modules. The stakeholders involved in this use scenario include the client’s decision maker and assistant located at the client’s office, on-site construction personnel, and related A/E and contractor engineers located off-site; these are represented by three dashed square areas in the figure. The blue arrows indicate data flows that support information sharing and collaboration during remote site visits, with the associated data types labeled alongside the arrows. The orange arrows represent the primary control command flows that coordinate site visits and image capture activities.

Figure 3. Data and control flows of the RASE system supporting collaborative remote site visits (Blue arrow: main data flow; Orange arrow: command control flow).

Regarding the primary data flow, on-site personnel capture live visual information using UAVs, smartphones, or 360° cameras and transmit video and voice streams to the Remote Collaboration Module through the manufacturers’ mobile applications installed on smartphones. These data are shared with the client and off-site engineers in real time and are further integrated with BIM information through the BIM/VR Integration Module. Real-time BIM visualization data, including viewpoint changes and newly created annotations, is exchanged bidirectionally between the BIM/VR Integration Module and the AR Module. This bidirectional exchange enables clients to observe current site conditions, compare them with design information, and record instructions directly on the corresponding BIM components. Meanwhile, off-site A/E and contractor engineers can simultaneously access both visual and BIM information through the Teamviewer-based teleconferencing environment and provide feedback during the visit, thereby avoiding delayed discussions and post-visit confirmations typical of conventional site visits.

2.1. Image Capture Module

Video recording and streaming devices are necessary for site personnel to record the site and stream it back to the client at remote locations. For image capture, we recommend mobile video capture devices that can be viewed and controlled wirelessly via smartphones or notebooks. Examples include smartphones and tablets, modern video cameras, UAVs, and 360-degree cameras that allow users to view and control them from a mobile device. The advantages are twofold. First, if the client only needs to view the site with a single view (e.g., a single video camera shooting the site), adopting a camera that can be viewed on a smartphone spares developers from learning and using the device manufacturers’ proprietary SDK. Second, since RASE’s AR module can accommodate and switch between multiple video recording devices, if the client prefers multiple views on-site (e.g., two video cameras or a UAV with a 360 camera), adopting a camera that can be viewed on iOS, Android, or Windows devices is mandatory for the switching functionality to work. This approach also allows the captured image on the mobile device to be instantly streamed wirelessly back to the client’s office without requiring additional storage, such as on a memory card or in the cloud.

UAVs can efficiently capture images of large areas or those at height from the sky, with less concern for the safety of video recording personnel. They benefit projects with large areas (e.g., highway construction, river reservation work) and those with hazardous or inaccessible conditions (e.g., high-rise buildings, offshore construction). Many modern UAV brands, such as DJI, Parrot, and Skydio, offer UAVs capable of autonomous cruise, long-range control, and streaming to mobile devices.

Nevertheless, a UAV has a limited communication range for control and image streaming. When the main visit area is inside a building under construction and complex partitions or obstacles are present, GPS may be blocked, and a UAV may not be a suitable tool for video capture. Therefore, regular mobile video cameras may still be needed in such scenarios as long as shooting can be conducted safely. Various types of video cameras are available on the market nowadays. Examples are smartphone cameras, compact cameras, DSLR cameras, and action cameras. All easy-to-carry cameras are suitable for RASE as long as they have remote streaming ability. The 360 cameras are also suitable because they can be attached to the workers’ safety helmets and capture 360-degree footage efficiently using back-to-back lenses. In this study, we used DJI Mavic 3 UAV and Insta 360 cameras for the video-capturing devices. Table 1 lists the major devices and tools used in the case study.

Table 1. Devices and tools used in the case study.

Types of Device/Software	Product Names Used by RASE
Image capture devices	DJI’s Mavic 3 (SZ DJI Technology Co., Ltd., Shenzhen, China) and Insta 360 One (Arashi Vision Inc., Shenzhen, China)
Image streaming portal	Two Android 7.0 smart phones
BIM software	Autodesk Revit (v2020.2.9), Naviswork (v2020)
VR/AR modeling software	Unity 3D (v2023.1.22f1) with C# (v. 9.0)
VR/AR device	HTC VIVE (High Tech Computer Corporation, Taoyuan, Taiwan)
Remote collaboration software	Teamviewer Remote (v15.39.5.0)
Computer	Asus Intel® Core i7-10750H CPU 2.60 GHz with Microsoft Windows 10 (ASUSTeK Computer Inc., Taipei, Taiwan)

2.2. Remote Collaboration Module

The remote collaboration module serves two purposes. First, it provides a portal to receive the image stream from different video recording devices. Second, it allows the client to control and adjust some functions of the video-capturing devices. This module relies on the screen-sharing ability of commercial online collaboration or video conferencing software such as ZOOM, Microsoft Teams, and Teamviewer.

This study chose Teamviewer because the software not only allows a user to see the shared screen of a remote device but also controls the device remotely. The image captured by either UAV or video cameras can be streamed to the smartphones on site and then shared with the client’s computer remotely via Teamviewer. The clients can also control and adjust some functions of the UAV or video cameras via Teamviewer.

2.3. BIM/VR Integration Module

The integration of the VR model in AR and the BIM model is necessary so that clients can view the mapping between the real world and the corresponding BIM components via AR goggles. BIM software, such as Autodesk’s Revit and Bentley’s Microstation, helps users build a 3D model of a building project. They are commonly used in the AEC industries for 3D design and construction management; however, they are not suitable for use directly with the VR model, which is required to realize AR capabilities. A common 3D file format between BIM software and VR development tools such as Unity and Unreal Engine 4 (UE4) is .fbx.

Thus, for compatibility reasons, this study suggests converting the BIM model to .fbx, which then can be read by VR development tools. Since the VR model will serve as the foundation for AR, a BIM model with complete texture information will certainly enhance clients’ immersive experience in AR. If the original BIM lacks texture information, the texture information must be defined in the VR model so clients can have an immersive, realistic experience.

To provide mapping between the BIM component and the real world in AR, appropriate zoning should also be determined, referred to as scenes in the following, and the BIM model’s components should be grouped accordingly. For example, in a building project, if clients prefer to visit the construction floor by floor, separating the BIM model into different floors is appropriate. It is easier to separate the 3D model in BIM software than in VR development tools because automatic filtering is possible based on component attributes in the BIM model.

Note that when importing the BIM model into Unity, we found that saving the model as .fbx and importing it directly can cause some building texture information to be lost or inadvertently changed. Also, the grouping of BIM components by zoning might be incorrect, with some components not placed in the subfolders to which they should belong. The problem was solved by first saving Revit’s file (.rvt) to Naviswork file format (.nwc), opening the model in Naviswork, saving it as a .fbx file, and then importing it to Unity. Therefore, we highly suggest using the Revit–Naviswork–Unity transition process to convert the BIM model to the Unity model instead of the Revit–Unity direct process. Figure 4 (left) compares categorization folders between the two file-converting processes, showing that, while the transition process correctly maintains the categorization of components by floors (e.g., FL2, FL3), several small member components (e.g., LED lights, pipe bends) are not put in appropriate folders if direct file conversion is used, as highlighted in red dash box. Figure 4 (right) shows that, while the texture of building exteriors is maintained using the transition conversion, texture information is lost using the direct conversion, as highlighted in the red dash box.

Figure 4. Comparison of transition and direct conversions.

2.4. AR Module

The purpose of the AR module in RASE is to present the corresponding VR model in the AR Head Mounted Display (HMD) the clients wear and allow them to:

• visualize the projected BIM model and compare it with the real site scene transmitted from image capture devices;
• switch between different image capture devices;
• switch between different BIM scenes;
• move in a scene;
• annotate text on components in a scene;
• update the BIM model;
• record the screen of AR’s HMD in action.

This module integrates all digital data sources and provides control functions to clients. It also requires the most coding work in RASE. This study used Unity and C# to develop the AR module because of the platform’s abundant development resources and online code examples. HTC’s Vive was also used as an AR HMD due to its familiarity.

2.4.1. User Interface

Figure 5 shows the three control buttons on an HTC controller that RASE used, namely, (1) trackpad, (2) system button, and (3) trigger. The following describes the AR user interface implementation.

Figure 5. Three buttons used by RASE: (1) trackpad, (2) system button, and (3) trigger.

Movement and Direction Turning

Like a first-person game, RASE allows the clients to be present in a virtual space with an immersive experience via HTC HMD. The clients may change their locations in the virtual world using the controller. RASE only provides the mapping between the real site and the virtual world at the scene level. Within the selected scene, the clients need to move to the location of the corresponding component by themselves if they need to compare it with the respective real scenes.

As shown on the left in Figure 6, pressing the center of the trackpad to pinpoint the desired location, then releasing it, will move one to that location. When the pointed location is marked with a highlighted red cross, the system does not accept the move because the location is illegitimate to move to (e.g., the column shown on the right in Figure 6).

Figure 6. Moving in the virtual space: legitimate move (left) and illegitimate move (right).

Since the HMD has sensors, clients can change their perspective by turning their heads. The virtual world’s perspective will be automatically adjusted accordingly. Sometimes, quick head-turning may cause the client discomfort. RASE also allows the clients to change their perspectives by pressing the up, left, right, and down parts of the trackpad to slow down the turning speed.

Illegitimate locations or movements are implemented using Box Collider, a built-in Unity object. Box Collider allows the developer to define the component or space as a solid entity that cannot be moved into. It can also be used to define the maximum height the clients can move on the ground or in an open space. Figure 7 shows an example of legitimate movement space, in which orange blocks are restricted areas that cannot be moved into.

Figure 7. Movement block areas (Orange rectangles: block areas with small boxes representing centers of the areas and borders).

Switching Between Real Scenes and the Virtual World

Clients may switch between real scenes streamed from on-site image-capture devices and the virtual world of the BIM model. The original function of the system button on the controller was changed to achieve this purpose. When the clients press the system button, a menu of currently available scenes is presented. The clients can then choose the scene they want to switch to. Figure 8 demonstrates switching between two real scenes, i.e., from a UAV for an exterior view and from a video camera for an interior view of the building.

Figure 8. Switching between different scenes.

Annotation

During the site visit, the clients may like to annotate comments on the components where questions or problems exist. The clients may press the trigger button to pinpoint the component in question. While pressing the trigger button, the clients may draw or write on the pointed components. Figure 9 shows two annotation examples. On the left, the client found a discrepancy between the opening location and the BIM model shown in Figure 8 shows two annotation examples: (1) the client wrote ‘opening’ (in Chinese) to indicate the incorrect opening location; (2) the client drew a ‘cross’ to indicate the absence of safety measure for the opening on the site.

Figure 9. Annotation examples.

Switching Scenes in the Virtual World

The BIM model contains a lot of information that can help clients navigate the virtual world efficiently. RASE provides a floating menu for each scene. Scene switching has two purposes: one is to allow clients to select the scene they want to visit, and the other is to control the types of components they want to focus on when they are in the selected scene. For example, the client may want to focus on building structure and, thus, disable the visibility of MEP components.

Figure 10 shows the design of the scene selection menu for the studied case. On the left is the menu that allows the client to select and switch the location (by floor) they like to visit. On the right is the sub-menu for component filtering, with scenes of different building components, such as MEP, columns, and beams.

Figure 10. Design of the scene switching menu for the case (Checking the boxes may include the selected the scenes and the corresponding components during the VR walkthrough).

The scenes by floor are arranged in the order of construction progress, namely, from the ground (0F) to the first floor (1F), second floor (2F), and eighth floor (8F). The clients may select the floors according to the project’s progress. Figure 11 shows the client used the trigger to point and press the buttons to change the scene. Appropriate selection of component types and exclusion of those not of interest reduces obstruction and complexity, making clients’ inspections easier. Figure 12 compares the scene with and without the MEP selected. Figure 13 shows an example of object settings and Unity scripts that change the scene when the trigger button is pressed.

Figure 11. Using the trigger to change the scene by mov the end point of direction line to the button and click the hand trigger.

Figure 12. Comparison of the scene with and without MEP.

Figure 13. Object settings and scripts for changing scenes.

Screen Recording

Recording the important screens during the visit is helpful for the client’s communication with engineers later on, regardless of whether the engineers responsible for the raised questions or problems are present. When the related engineers are present, the recording serves as a checklist to help them ensure they go through and resolve all the client’s problems afterward. When the engineers are absent, the recording is even more helpful and may help the on-site person communicate with the related engineers after the site visit.

RASE can record the client’s visit from the client’s perspective and save it as a 3D panoramic video clip. The panoramic video is better than the 2D video or still image, in which the view angle is fixed. The panoramic video allows engineers to explore not only the location in question but also its surrounding area from different angles. We used the AV_Movice_Capture object provided by Unity to implement the scene recording module. Figure 14 shows: (1) the panoramic screen capture clips recorded and stored in the designated folder and that (2) engineers may click and view the clip from a different angle.

Figure 14. Screen recording module.

2.4.2. System Development Summary

Since the development of the module above requires using SDKs (Software Development Kits) from different sources, Table 2 lists the main functions, their usage, and the SDK packages used for the development. Note that the buttons on both right- and left-hand controllers are equipped with all implemented functions. When the controller is referenced in the table, the client can execute the operation using either controller.

Table 2. Resources used for the development of RASE.

RASE’s Function	Operation Method	Development Resources
Changing view angle	Directly turning the head (i.e., HMD). Press up, down, right, left of the controller’s trackpad (right- or left-controller).	Unity’s XRInteraction object Vive’s built-in modules Self-coding modules
On/off and switching site view	Press the system button of either right- or left-controller. Use the trigger to point and select the scene from the pop-up menu for scene switching (right- or left-controller).	Vive’s built-in modules Self-coding modules
Movement	Press the trackpad’s center, point to the new location to move to, and release the button to move to the new location (right- or left-controller).	Unity’s XRInteraction object Self-coding modules
Menu item selection	Point to the menu item and press the trigger to select the menu item (right controller). e.g., used when switching scene and filtering components.	Self-coding modules
Annotation	Point to the component to be annotated and press the trigger to start drawing (left controller).	Unity’s XRInteraction object Self-coding modules
Screen recording	Press the space bar of the computer to toggle the screen recording.	Unity’s XRInteraction object Self-coding modules

3. Benefit Evaluation

In this case study, only the structural and MEP components of the building project were implemented and evaluated. Construction items such as site work, tiling, windows, and interior finishes were not included in the analysis. It is assumed that a project BIM model is already available, and the setup and learning time required for RASE are not included, as these durations vary with the client assistant’s familiarity with software tools such as Unity, TeamViewer, and RASE. All estimated time and cost values are assumed to occur at the end of each month throughout the project duration.

It should be noted that the purpose of this evaluation is not to generalize or advocate specific percentages of time and cost savings, as these outcomes are highly sensitive to multiple factors, including project size, frequency of routine site visits, site accessibility, and travel distance and transportation costs between the client’s office and the construction site. Rather, the evaluation aims to identify threshold conditions under which the proposed RASE approach becomes advantageous in terms of time and cost savings for the studied case.

Figure 15 compares the workflow of a conventional client routine site visit with that of a remote site visit using the proposed RASE framework. It is assumed that a BIM model has been established and is regularly updated, and that the client intends to record revision instructions by annotating the corresponding BIM components. Owing to the site’s remoteness, the conventional visit requires the physical presence of the client, the A/E project manager, and the contractor’s site personnel, while engineers responsible for specific components may only participate remotely because which component might have an issue is not known before the visit.

Figure 15. Comparison of workflows between a conventional client routine site visit and a remote site visit using the RASE framework.

The two approaches differ primarily in data preparation, transportation requirements, the timing and location of BIM annotation, as well as the mode of technical discussion. In the conventional approach, data preparation is often substantial, particularly when BIM is not readily available. Even when BIM is in place, on-site access requires either transporting computers or laptops or relying on cloud-based BIM platforms with sufficient wireless bandwidth and storage space. In contrast, the RASE approach requires an initial investment in AR equipment and system setup, including categorizing BIM components for site-visit purposes. Once deployed, however, RASE significantly reduces per-visit preparation effort and wireless data transfer, as BIM and the RASE system reside on the client’s computer and only real-time video streams are transmitted, which can be adequately supported by current 4G networks.

As illustrated by the solid arrows in Figure 15, conventional site visits involve sequential physical movement between areas, with clients making memos and recording observations while accompanied by on-site personnel. BIM annotation and documentation are typically completed after returning to the office. By contrast, RASE enables clients to communicate with on-site personnel and the engineers responsible for the raised issue, but not on site, via teleconferencing, to annotate instructions directly on BIM components in real time, and to switch between predefined scenes within the mixed-reality environment without physical relocation. This fundamentally alters the visit from a location-dependent process to a digitally coordinated one.

The dashed arrows in Figure 15 further highlight differences in technical coordination. In the conventional approach, unresolved questions are commonly deferred and later discussed with in-house engineers at the A/E or contractor’s office, introducing delays and potential information loss. With RASE, responsible engineers can participate in the same remote session, allowing issues to be addressed immediately and reducing the need for follow-up communication.

Overall, RASE eliminates transportation between the client’s office and the site, yielding clear time and cost savings that depend on visit frequency, travel distance, and the number of participants. While the duration of on-site inspection activities remains similar in both approaches, RASE shortens the overall decision cycle by enabling real-time BIM annotation and synchronous collaboration. Consequently, RASE transforms site visits from a sequential, travel-intensive process into a more integrated, data-driven, and collaborative decision-support activity. In addition, a remote site visit may pose less of a safety concern to clients, since they may not be as familiar with the dynamic construction environment as construction professionals.

The following further analyzes the benefits of using RASE in detail from the perspectives of time, cost, and safety based on the studied case.

3.1. Time Saving

This section estimates the time the client saves with RASE compared to conventional site visits, using the studied case. Assuming the project BIM model is available, Table 3 shows the case parameters for estimation. Table 4 shows the calculation for estimating time savings. The total estimated time for the concerned site-visit activities is 7174 min for the conventional on-site visit and 1576 min for the RASE approach, resulting in 78% savings. Note that the estimate focuses on the per-visit saving and does not consider the learning and setup time for RASE.

Table 3. Case project information.

Project Parameters	Value	Project Parameters	Value
Project location	Hsinchu	Visit times per month	2 times/month 1 time/month for the last two months
Client’s location	Taipei	RASE’s BIM conversion time	3 workdays
Building size	12 stories with 1 basement	Memo time per visit	30 min
Project duration	24 months
Transportation time	64 min
Transportation cost	NT$580

Table 4. Estimation of time saving.

Work Item	Conventional	RASE	Saving
Preparation time per project
Transportation	2944 min	0 min	2944 min
	(64 × 2 × 22 + 64 × 2 = 2944)
RASE setup	0 min	1440 min	−1440 min
		(3 × 8 × 60 = 1440)
Site visit time per floor
Checking project progress	10 min	10 min	0 min
Check discrepancies	30 min	30 min	0 min
Check deficiencies	20 min	20 min	0 min
Check occupational safety and health	30 min	30 min	0 min
Memo time per project
Organize records after visit	1380 min(30 × 2 × 22 + 30 × 1 × 2 = 1380)	0 min	1380 min
Storing and sending data	2760 min(60 × 2 × 22 + 60 × 1 × 2 = 2760)	46 min(1 × 2 × 22 + 1 × 1 × 2 = 46)	2714 min
Total time per project
Organize records after visit	7174 min(2944 + 10 + 30 + 20 + 30 + 1380 + 2760 = 7174)	1576 min(1440 + 10 + 30 + 20 + 30 + 46 = 1576)	5598 min

Figure 16 compares the estimated incremental time spent as the project progresses. This indicates that at the 4.5th month, RASE’s approach began to save time compared to the conventional visit, resulting in a 78.0% saving in the 24th month. Note that an additional green number series estimates the time spent on the on-site visit using a UAV, which is not presented in the table.

Figure 16. Comparison of time saving among conventional and RASE approaches.

3.2. Cost Saving

Table 5 estimates the cost savings by clients using RASE compared to conventional site visits based on the case. The cost of the conventional visit is mainly the transportation fee and personnel travel time, while RASE’s cost is the initial purchase of a VR, a UAV, and a 360-degree camera. The result shows that the cost savings from a single case, such as the one surveyed, did not offset the equipment cost incurred by RASE. The RASE approach costs 44.1% more than the conventional site visit, as shown in Figure 17. Nevertheless, the equipment is a one-time purchase that can be used for a long time and will eventually pay off for subsequent, larger, or more projects. Note that the cost estimate for the conventional visit assumed the use of an existing smartphone to record images and did not include any equipment purchase cost.

Table 5. Estimation of cost saving.

Item	Conventional	RASE	Saving
Transportation	NT$26,680 (=580 × 2 × 22 + 580 × 2)	0	NT$26,680
RASE setup		NT$85,000 (=35,000 (VR) + 15,000 (Insta360) + 35,000 (UAV))	−NT$85,000
Personnel time cost	NT$24,030 (=(7209/60) × 200)	NT$5370 (=(1611/60) × 200)	NT$18,660
Total	NT$50,710(=26,680 + 24,030)	NT$90,370(=85,000 + 5370)	−NT$39,660

Figure 17. Comparison of cost among conventional approaches and RASE.

3.3. Safety Perspective

From a safety perspective, RASE reduces the client’s risk of visiting the site overall by reducing the number of physical visits to the construction site. The client’s representatives, as well as the architect, contractor’s executive managers, inspection facilitator engineers, and superintendents, are usually present in the conventional site visit. Comparatively, the clients’ representatives usually have less construction site experience, and they may be unfamiliar with the dynamically changing site and are therefore riskier on site (e.g., slipping, tripping), especially during visits at height.

In addition to the inspection engineers, facilitators responsible for taking photos and memos are also necessary for appropriate documentation for the follow-up after the visits. RASE’s use of UAVs reduces the risk of capturing images from height. Attaching a 360-degree camera to a facilitator’s helmet lets him watch each step while keeping both hands free while patrolling the site.

3.4. Sensitivity Analysis

Previous analysis shows that RASE saved time but did not reduce costs in the studied case due to the initial cost of purchasing the necessary equipment. However, in the long run, when the purchased equipment is used in multiple projects, RASE’s approach will eventually save cost compared to the conventional approach.

Nevertheless, some clients may not have a second project in a short period of time. Therefore, this subsection further analyzes cost savings for projects of different sizes and distances to determine under what circumstances RASE can reduce costs for a single project.

• Variation of Distance

Suppose the client’s office is in Hsinchu, and the same project, originally located in nearby Taipei, is now considered to be placed at different, farther locations, namely, Kaohsiung and Taitung. Both cities are far from the client’s office, but Kaohsiung can be reached via a high-speed train (NT$2400 and 444 min per round trip), which takes much less travel time than Taitung (NT$3144 and 168 min per round trip). Table 6 compares the total time and cost savings for the entire project duration at the three locations. Figure 18 graphically compares the time and cost savings for the Taipei, Kaohsiung, and Taitung locations. The RASE’s approach saves tremendous time and costs for both the Kaohsiung and Taitung locations. Note that in this case study, what matters is the transportation time and cost from the client’s office to the site, not the distance in between.

Table 6. Estimation of time and cost saving for different project locations.

Project Location	Conventional	RASE	Saving
Hsinchu–Taipei	7174 min NT$50,710	1576 min NT$90,370	5598 min −NT$39,600
Hsinchu–Kaohsiung	11,958 min NT$150,377	1576 min NT$90,370	10,382 min NT$60,007
Hsinchu–Taitung	24,654 min NT$226,921	1576 min NT$90,370	23,078 min NT$136,551

Figure 18. Comparison of savings on time (a) and cost (b) for Taipei, Kaohsiung, and Taitung locations.

• Variation of Visit Frequency

As shown previously, the Taipei case is the nearest location to the client’s office and may save time (79.0%) with RASE, but not the cost (44.1% extra), assuming a visit to the site twice a month. As shown in Figure 19, increasing the visiting frequency to four times per month reduces time and cost savings to 89.2% and 10.4%, respectively. Increasing the visit frequency to six times per month further increases savings in time and cost to 92.5% and 67.1%, respectively. Thus, the frequency of visits has a significant effect on time and cost savings for this two-year project.

Figure 19. Comparison of savings on time (a) and cost (b) for different visiting frequencies.

• Variation of Project Size

As shown previously, the Taipei case is the nearest location to the client’s office and may save time (79.0%) with RASE but not the cost (44.1% extra), with the project size of 12 stories (24-month duration with 2 visits per month). As shown in Figure 20, increasing the project size to 18 (30-month duration) and 24 stories (36-month duration) yields time savings of 83.2% and 86.1%, respectively. However, the increased project duration does not offset the initial purchase cost of RASE equipment. Nevertheless, it reduces the extra cost from 44.1% to 29.5% and 14.5%, respectively.

Figure 20. Comparison of savings on time (a) and cost (b) for different project durations.

4. Discussion and Future Research

4.1. Advantages and Disadvantages

From the clients’ perspective, compared with a conventional physical site visit, the RASE approach has the following advantages and disadvantages:

• The RASE approach provides the clients with more flexible but limited mobility.

Using RASE with a variety of image tools, such as UAVs, 360-degree cameras, or sports cameras, provides more flexible mobility for site visits by allowing clients to remotely see the constructed components in dynamic, unsafe, or high-elevation areas of a construction site without the risk of a site visit. However, when the visiting site is static and safe, a physical site visit may offer faster, more convenient mobility with easier human communication than the RASE approach.

• The RASE approach provides clients with a safer, cost-saving, and time-saving visit, though that is not always the case.

The RASE approach may always provide clients with a safer site visit because they do not need to be on-site, and thus there is no exposure to on-site risk. However, the RASE approach does not always save time and cost when the deployment time and cost are considered. The benefits saved depend on the number of required visits, the distance between the client’s office and the site, transportation costs, and time.

• The RASE is most beneficial for remote projects that need to integrate site-visit records with specific BIM components.

Physical site visit records may consist of different photos and videos (e.g., from UAV, smartphone, digital camera, 360 camera) as well as clients’ instructions and contractor’s responses. The instructions and responses are easier for engineers who were not present during the site visit to understand and trace if the instructions are annotated with the specific components in the BIM. The RASE approach provides real-time, instant association rather than requiring later off-site steps.

4.2. Future Research

The preceding discussion indicates that the RASE approach is most beneficial when a BIM model is already available and when clients need to record revision instructions and manage change orders in a structured, traceable manner. From a broader research perspective, this highlights the opportunity to advance RASE from a human-operated integration framework toward a more intelligent, data-driven system that supports multi-source data fusion and collaborative decision making in dynamic construction environments. Future research directions are outlined as follows:

• Automatic recognition and association of multi-source visual data.

In its current form, RASE enables clients to manually select areas or components and associate visual streams from UAVs, smartphones, and 360-degree cameras with corresponding BIM scenes. Future research should investigate computer vision and image recognition techniques to automate this association process. By combining visual recognition with positioning technologies (e.g., GPS, visual-inertial odometry, or indoor localization), images captured by handheld or wearable devices could be automatically mapped to relevant BIM components. This capability would reduce manual switching and cognitive load during remote visits. UAV-based imagery presents additional challenges due to its wide coverage and mixed content (e.g., progress overview versus task-specific views), requiring further research into scale-aware and context-sensitive association methods.

• Voice-based interaction and semantic input for decision support.

Textual input and drawing in VR/AR environments remain less efficient than desktop-based interaction. With recent advances in speech recognition and large language models, voice-based interaction has become a viable alternative. Future research may explore voice-driven system navigation, component selection, and semantic annotation, allowing clients to issue commands, record instructions, and generate BIM-linked annotations through natural language. Such capabilities would enhance usability and support more fluid collaboration between remote clients and on-site personnel.

• Event-driven alerts and real-time decision assistance.

Another important research direction involves integrating automated event-recognition techniques, including progress deviation detection, safety-risk identification, and abnormal activity recognition. Embedding these capabilities into RASE would allow the system to proactively alert clients during VR/AR navigation and guide their attention to potential problem areas. Rather than replacing human judgment, these event-driven cues could function as decision aids that improve situational awareness and support timely intervention.

• AI-enabled orchestration of collaborative data capture and analysis.

The above directions primarily automate individual steps of the remote-site-visit process, which risks creating isolated “islands of automation.” Future research should explore AI-enabled orchestration mechanisms that coordinate data-capture devices, BIM environments, and human interactions within a unified workflow. For example, based on project progress and detected anomalies, the system could suggest or automatically plan an optimal sequence of areas to visit. Upon client confirmation, UAVs, cameras (held by on-site personnel or robots), and VR/AR interfaces could automatically and collaboratively focus on identified problem areas to collect additional evidence. Human-in-the-loop confirmation would remain central, ensuring that automation enhances—rather than replaces—collaborative decision making.

5. Conclusions

Although construction site inspection at the activity and operation levels still requires physical presence, remote site visits can be both feasible and valuable for project clients, who typically focus on project-level progress, deviations, and decision making. This study proposed an innovative framework and developed the RASE system to support remote client visits by integrating UAVs, 360° cameras, BIM, and VR/AR technologies. RASE enables clients to visualize and synchronize real site conditions with BIM models at predefined scene levels, annotate issues directly on corresponding components, and flexibly switch among heterogeneous visual data sources. By leveraging existing desktop-sharing technologies, the framework achieves these functions with relatively low development complexity.

The case study of a 12-story building project demonstrates that RASE can reduce site-visit time by 78.0%, although total costs increased by 44.1% due to initial equipment investment. Importantly, RASE significantly reduces on-site risk for clients. Sensitivity analyses further indicate that greater travel distance and higher visit frequency substantially improve both time and cost effectiveness, whereas longer project duration has a comparatively smaller impact. These findings suggest that while RASE may not always reduce costs in single, nearby projects, it consistently delivers time savings and safety benefits and becomes increasingly advantageous for remote projects or frequent site visits.

Beyond system implementation, this research contributes to the scientific understanding of remote construction monitoring by framing site visits as a problem of multi-source data fusion and collaborative decision making. The results show that interaction design and data synchronization are critical to enabling shared situational awareness, and that BIM can serve as a semantic anchor, linking heterogeneous visual data and human communication in real time. This perspective advances BIM from a static information repository to an active coordination medium, aligning with emerging digital-twin research while emphasizing human-in-the-loop decision processes.

Looking forward, the RASE framework can be extended toward intelligent, human-centered automation by integrating computer vision, localization, speech-based interaction, and semantic BIM linkage. Such developments would allow remote site visits to evolve from manual visualization toward proactive, event-driven decision support. AI-enabled coordination of sensing, visualization, and interaction workflows has the potential to further enhance digital-twin–based construction management while preserving essential human oversight in dynamic and high-risk environments.