The New Norm for Construction DX: Smart Sites Realized by Gaussian Splatting × LRTK

In the construction industry, digital transformation (DX) has been progressing in earnest in recent years, accelerating a shift from traditional two-dimensional data such as drawings and photos to visualization using three-dimensional data. Capturing and sharing a site in 3D makes it easier to convey the finished image intuitively, discover design mistakes, and align understanding among stakeholders. As a result, 3D utilization that improves quality, safety, and efficiency is becoming a foundational technology for construction DX.

One technology attracting attention in this trend is Gaussian Splatting. At the same time, a solution called LRTK has emerged that enables easy measurement of high-precision point clouds using a smartphone, and its adoption is expanding. In this article, we gently explain what Gaussian Splatting is and why it is gaining attention in construction DX now. We also describe the characteristics of point cloud data obtainable with LRTK, the benefits of combining Gaussian Splatting with LRTK, concrete on-site use cases (construction management, consensus building, inspection, etc.), the appeal of this technology’s visualization accuracy and lightweight rendering, and future prospects. Finally, we explain why you should start acquiring spatial information with LRTK now, looking ahead to the future. The tone is accessible to non-technical readers, so please use this as a reference for creating future smart sites.

What is Gaussian Splatting?

Gaussian Splatting (often abbreviated 3DGS) is a cutting-edge 3D representation technique that reconstructs a three-dimensional scene from multiple photographs and can rapidly generate images from novel viewpoints. Its distinguishing feature is that it does not convert a scene’s volumetric information into surfaces such as polygons; instead, it projects and composes countless elements represented by Gaussian distributions directly into space and renders them. In other words, it can be seen as a type of point cloud method that treats each point as a softly spread “3D Gaussian” (a three-dimensional Gaussian distribution).

Specifically, features are detected from multiple input-view images to create a point cloud. For each point, surrounding color and brightness information are blurred and averaged using a Gaussian function, and colored elliptical points are placed in 3D space as if splattering paint on a canvas. By layering tens of thousands of these “Gaussian-blurred points,” a smooth, realistic 3D space is recreated, as if photographed. Because the spaces between points naturally blend rather than forming the blocky faces of traditional polygon models, it can produce photorealistic representations that are pleasing to the eye.

What makes Gaussian Splatting revolutionary is its fidelity and processing speed. For example, it can capture textures that were difficult to express with conventional 3D CG—such as the transparency of water, reflections in glass, and metallic sheen—and turn them into 3D data. It can also represent highly detailed features like many overlapping leaves or individual animal hairs. Moreover, a major attraction is that such high-detail models can be generated and displayed in relatively short times. In research examples, high-quality 3D models were produced from about one minute of video capture followed by roughly 30 minutes of machine learning processing. Recently, simple apps that complete capture-to-3D-model pipelines on smartphones have appeared, enabling users to confirm results on the spot with 1–2 minutes of shooting and several minutes to tens of minutes of computation.

This technology belongs to the lineage of AI methods for image-to-3D reconstruction that emerged around 2020, notably NeRF (Neural Radiance Fields). NeRF was revolutionary but required long model training times, making real-time use difficult. Gaussian Splatting represents scenes explicitly with Gaussian-shaped point clouds without relying on neural networks, achieving both extremely fast processing and high rendering quality. As a result, the real-time rendering methods announced in 2023 shocked the industry and have rapidly spurred applied research and demonstrations in many fields, including construction.

Why is it attracting attention in construction DX now?

There are two main reasons Gaussian Splatting is gaining attention in construction: the increasing necessity of adopting digital technologies, and the practical evolution of 3D technologies.

From an industry-wide perspective, the Ministry of Land, Infrastructure, Transport and Tourism began applying BIM/CIM (use of 3D models) as a principle to some public works from fiscal 2023, making 3D data utilization effectively essential. Additionally, labor shortages and an aging skilled workforce have made digitalization and automation urgent for work efficiency and knowledge transfer. Relying on paper drawings and 2D photos makes it increasingly difficult to keep complex projects within schedule and budget; improving productivity through DX is unavoidable. By utilizing 3D site data, it has become clear that efficiency can be achieved consistently from planning through construction, inspection, and maintenance. For example, using point cloud measurement data of as-built conditions after construction to check component dimensions has been reported to reduce work time and cost by about 73% compared to traditional manual measurement. If scanned 3D data automatically highlights differences with color coding, millimeter-level errors that human eyes might miss can be detected, greatly contributing to quality assurance and reducing rework. In this way, 3D visualization has become an indispensable element of a productivity revolution on site.

On the technological side, not long ago, recording a site in 3D required expensive laser scanners or contracting specialized vendors. But now, we have arrived at an era where anyone can acquire 3D point clouds using smartphones or drones. For example, combining LiDAR sensors or high-performance cameras on the latest smartphones with RTK-GNSS receivers enables collection of high-precision point cloud data simply by walking around a site. LRTK condenses these technologies into a solution that allows anyone to obtain point clouds with absolute coordinates with just one hand, even without surveying expertise. Moreover, when combined with AI technologies like Gaussian Splatting, it has become realistic to automatically generate detailed 3D models from captured photo sets. Tasks that used to take hours in photogrammetry software can be completed much faster with Gaussian Splatting, yielding lightweight, easy-to-handle models. In other words, the barriers to acquiring and utilizing 3D data have dramatically lowered—and that is why “now” is the time. Adopting DX at this timing offers the chance to maximize the benefits of the latest technologies while fundamentally streamlining business processes.

Characteristics of point clouds and spatial information obtainable by LRTK

LRTK is a solution that uses a high-precision GNSS (RTK method) receiver attached to a smartphone, enabling anyone to easily measure three-dimensional point cloud data. A point cloud is 3D data representing objects or terrain as a collection of countless points, each of which can include positional coordinates (X, Y, Z) and, in some cases, color information (RGB) or reflectivity. The greater the number of points (higher density), the more finely the shape can be reproduced—similar to how pixels form an image. LRTK uses a smartphone’s LiDAR sensor or camera to scan the surroundings while simultaneously using RTK-GNSS to determine the measurer’s position with centimeter accuracy (half-inch accuracy), thereby acquiring point clouds whose points have absolute coordinates (latitude, longitude, altitude).

There are several important characteristics of data obtained by this “point cloud scan × RTK positioning” approach. First, because every point has geographic coordinates (such as the World Geodetic System), the acquired point cloud can be overlaid directly onto maps or design coordinate systems. Traditionally, point clouds from laser scanners or smartphone LiDAR were highly accurate in relative site geometry but required separate target placement or post-processing alignment to fit into public coordinate systems. With LRTK, you can obtain 3D data already tied to global coordinates, dramatically simplifying comparisons between site surveys and drawings/BIM data. For example, when combining point clouds measured on different days, each dataset aligns precisely on a common coordinate system, making it easy to create integrated 3D models of wide-area terrain or large structures. Furthermore, RTK positioning provides very high accuracy, typically within a few centimeters (within a few in), so measurements of distance, area, and volume on the resulting point cloud model yield precision comparable to actual field measurements. In other words, LRTK fuses 3D scanning and positioning to generate digital spatial information that is “true-to-scale and high-precision.”

LRTK-acquired point clouds can also be augmented with color information from photographs. By using images taken with a smartphone camera to texture-map the point cloud, the result becomes rich 3D data that retains the actual colors of the site rather than a monochrome collection of points. Such color point clouds and 3D mesh models automatically generated from point clouds can be easily viewed and shared in the cloud without specialized software. LRTK allows measurement data to be uploaded to a dedicated cloud, shared via a web-browser 3D viewer with stakeholders, and provides functions for immediate measurement of distance, cross-sections, and volume. The ability to check collected site information from the office, instantly perform quantity calculations for orders and construction management, and otherwise support everyday use of point cloud data is a core strength of LRTK.

Gaussian Splatting × LRTK: What happens when they are combined?

So, what happens when the Gaussian Splatting introduced so far is combined with 3D point clouds obtained by LRTK? In short, you get a “smart site” that digitally reproduces physical space almost exactly as it is. The photorealistic expression strengths of Gaussian Splatting combined with the high-precision positioning information of LRTK point clouds enable construction of a digital twin that offers both precision and visual clarity.

Historically, there has been a trade-off: measurement-oriented point clouds were accurate in shape but coarse in appearance, while photo-based 3D models looked realistic but lacked dimensional reliability. Combining the two yields data that captures the strengths of both. LRTK point clouds provide the skeleton (precise geometric information), while Gaussian Splatting contributes the facial features (detailed color and texture). For example, you can measure spatial dimensions accurately from arbitrary viewpoints using point cloud data, while simultaneously inspecting those same locations with Gaussian Splatting’s high-fidelity visuals—enabling the observation of fine concrete cracks or reading the text on labels attached to equipment. Micro-details that are hard to distinguish from the point cloud alone are complemented by the overlaid photographic texture, allowing the site to be reproduced in data at a level indistinguishable from being there in person.

In other words, you obtain next-generation site data that unifies precise measurement data and immersive visuals. This combination, which balances measurement, recording, and visualization at high levels, could become the new norm in construction DX.

On-site use cases and future outlook

Smart site data produced by Gaussian Splatting and LRTK is expected to be useful across many construction tasks. Below are examples in construction management, consensus building, and inspection/maintenance, along with their potential benefits.

• Construction management: Recording and sharing progress with 3D models streamlines verification of as-built conditions and quality control. For example, scanning a structure immediately after construction with LRTK and comparing the resulting point cloud + GS model to the design BIM model can immediately detect shape deviations or construction errors. Tasks that traditionally involved cross-checking section drawings and site photos can be performed intuitively on a digital twin, reducing rework. Remote managers can also view the site model from the office and issue instructions easily. Visualizing daily progress in 3D fills the information gap between site and office and accelerates decision-making.

• Consensus building: 3D utilization is effective for communication with clients and local residents. For example, visualizing the pre-construction environment realistically with a GS-augmented point cloud and compositing a planned building model into it intuitively conveys the post-construction image. Scale, which is often hard to communicate with drawings or CG perspectives alone, becomes much more persuasive when integrated with the actual surrounding scenery in a 3D model. Stakeholders can discuss while experiencing a virtual site tour, facilitating agreement on design changes or local briefings. In the future, this technology could be used with AR glasses to overlay completion predictions on-site for real-time experience.

• Inspection and maintenance: GS × LRTK data is powerful for infrastructure and building maintenance. Periodic 3D scans of structures allow accumulation and comparison of aging changes digitally. For example, you can quantitatively assess whether bridge cracks have widened compared to the previous year or how equipment in a tunnel has changed, by comparing models. Because point clouds include position information, abnormal areas can be pinpointed accurately on the actual structure to aid repair planning. High or confined-area inspections can be performed safely without human entry using drone × LRTK-acquired GS models. In the future, applications may expand to AI-based automatic crack detection and degradation prediction using multi-year scan data.

These are just examples, but the realistic 3D site data created by Gaussian Splatting and LRTK will bring new value to every stage of construction projects. As visualization accuracy and shareability improve, all stakeholders can discuss and decide while looking at a common virtual reality, which is expected to raise overall project productivity and transparency.

The appeal of Gaussian Splatting’s visualization accuracy and low-load rendering

Another reason Gaussian Splatting is attracting attention is that it combines high visualization accuracy with low rendering load. As mentioned earlier, the technique overlays photographic information as point-like elements; because each point has a soft spread, overall realism is not greatly degraded even if the number of points is reduced somewhat. Traditionally, achieving high-detail 3D modeling required handling enormous numbers of polygons or points, leading to massive data sizes and rendering loads. However, in Gaussian Splatting each point can represent information over a certain area, allowing smooth appearance with relatively small data volume. In practice, by cleverly adjusting per-point color, opacity, and shape and layering them, it has succeeded in generating imagery close to real photographs in real time.

This lightness is a major advantage for on-site data use. For example, if high-detail textured 3D models can be displayed in a web browser rather than a specialized viewer, anyone with a PC or tablet can check site data. Gaussian Splatting is well suited to such use cases, and browser-based GS viewers have appeared in recent years. Without installing dedicated software, users can view 3D site models simply by clicking a URL, which greatly smooths internal and external information sharing. On the distribution side, a new compressed format optimized for Gaussian Splatting (SPZ format) has been proposed, enabling maintenance of high quality with lightweight file sizes. These trends show that Gaussian Splatting is not just photorealistic but also realizes “easy-to-handle 3D data.”

Low rendering load also broadens the potential for future real-time use. Currently, real-time display at over 30 FPS has been reported, and it is conceivable that scanning a site and instantly displaying a 3D model could become feasible. For example, if a GS model could be generated and shared to the cloud concurrently while walking a construction site during work, as-built checks and remote support could be performed on the spot. Because Gaussian Splatting balances visualization quality and speed, it is expected to become increasingly integrated into practical workflows.

Future R&D and expected application areas

The potential unlocked by combining Gaussian Splatting with LRTK will continue to expand. From a research perspective, dynamic scene capture and enhanced real-time processing are promising directions. Currently, reproducing static sites is the primary use, but in the future it may be possible to record and replay construction progress along the time axis as a four-dimensional model (4D Gaussian Splatting), or to capture the movements of heavy equipment and workers in real time for safety management. Additionally, analyzing captured high-fidelity data with AI to automatically detect changes and anomalies—a “smart inspection”—is a promising application. Image-based AI for detecting cracks and corrosion has advanced, and combining it with GS would enable accurate spatial localization in 3D and quantitative evaluation of degradation over time. As R&D progresses, the accuracy of such automated modeling and analysis is expected to improve dramatically, further accelerating construction DX.

Applications will also spread beyond construction. Gaussian Splatting is already being demonstrated in general contractors, shipbuilding, infrastructure maintenance, cultural heritage archiving, and education. In cultural heritage, for instance, its high fidelity and speed are invaluable for digitally preserving historic structures or creating VR exhibits in museums. Urban planning is another area where rapid generation of neighborhood-scale 3D models from drone imagery can be used for sunlight simulations and evacuation route planning. In contexts such as autonomous construction machinery and robotic construction, realistic digitalization of site environments is essential; high-precision site models created by GS × LRTK could be used for simulation training so AI robots can perform tasks accurately. In entertainment, the technique is expected to be used to incorporate real landscapes into games and film production. Because it enables scanning the real world into virtual spaces, it is highly compatible with the metaverse and VR/AR content and presents broad business opportunities.

Thus, the fusion of Gaussian Splatting and precise spatial measurement technologies will produce innovative solutions across diverse fields. For those of us involved in construction DX, it is important to keep up with related technological trends and quickly incorporate them into internal business reforms.

Closing: Why you should start acquiring spatial information with LRTK now

The future smart-site vision drawn by Gaussian Splatting and LRTK is not distant science fiction but already near. You may think, “Isn’t it still too early to use these technologies in-house?” However, history shows that earlier adoption of DX trends is advantageous. If you begin acquiring and accumulating high-precision spatial information now, you will gain a significant advantage when these technologies become industry standards.

First, data is a new asset. By archiving site 3D point clouds and models, you create a foundation for many future analyses and applications. If you later want to view past site conditions in 3D but have no data, nothing can be done. LRTK makes it easy to digitally archive current site conditions, which will be a valuable asset when reviewing past data for renovations or inspections.

Second, expertise in site DX cannot be accumulated overnight. Workflows for 3D scanning and data utilization need to be optimized through actual on-site practice. Starting point cloud measurement with LRTK now will build in-house know-how so that when the technology matures, you can utilize it more effectively than competitors. Conversely, delayed adoption risks falling behind when others have normalized 3D utilization and you lack the necessary expertise.

Third, the immediate benefits today are substantial. LRTK’s simple point cloud scanning is an investment for the future and already delivers tangible improvements in surveying, drawing creation efficiency, and sophisticated site records. As mentioned earlier, even using point clouds for as-built management can significantly reduce time and improve quality. Thus, rather than “enduring now for a future payoff,” you can adopt LRTK because it is useful now—creating a virtuous cycle that also prepares you for the future.

Fortunately, tools like LRTK have made what once only specialists could handle—high-precision surveying and point cloud technology—accessible to anyone on site. The fact that you only need a smartphone to get started makes it approachable even for DX beginners. Taking a small first step now is key. In preparation for the new norm of site digital twinization, consider beginning spatial information acquisition with LRTK today. The future of smart site creation has already begun.