Why "Gaussian Splatting" Now? The Overwhelming Realism of Digital Twins Enabled by LRTK

In recent years, "digital twins" that faithfully reproduce on-site conditions in virtual space have attracted attention in the surveying, construction, and CG fields. Using point cloud data acquired by drones or laser scanners and 3D models reconstructed by photogrammetry, there are active efforts to inspect sites remotely and use them for construction planning. However, conventional 3D models have had limits in reproducing reality in terms of detail and smoothness; especially with point clouds and polygon meshes, achieving a "being there" level of realism has required advanced processing or very large datasets.

Against this backdrop, a new technique called "Gaussian Splatting" has emerged. This is a 3D reconstruction method born from recent CG research that draws attention for its ability to render photo-like realistic 3D scenes quickly by smoothly layering vast numbers of points. It has begun to be incorporated into the latest cloud services and software, revealing the possibility that anyone could generate digital twins with overwhelming realism using data from drone shoots or smartphone measurements.

At the same time, the acquisition of on-site 3D data—the basis of those digital twins—has also advanced significantly. High-precision 3D measurement that once required specialized equipment can now be easily performed by site personnel themselves using smartphone devices combined with RTK-GNSS—so-called LRTK devices. LRTK lets you collect point cloud data with high-precision position information on a smartphone, and combined with the ultra-photorealistic visualization enabled by Gaussian Splatting, this pairing makes it possible to digitally reproduce and share the site's "now" with unprecedented speed and accuracy.

This article first explains what Gaussian Splatting is, its technical background, and how it differs from point clouds and meshes in an easy-to-understand way. It then introduces the significance of applying this technology to digital twins (site fidelity, lightweightness, smoothness, rendering speed, etc.), touches on compatibility with LRTK and the benefits expected from adoption, and considers concrete workflows for generating visual models by applying Gaussian Splatting to point clouds acquired with LRTK. It also discusses future application prospects such as integration with construction DX and BIM/CIM, cloud sharing, and web display. Finally, it summarizes how introducing LRTK can help leverage these cutting-edge technologies on-site.

What is Gaussian Splatting? Differences from Point Clouds and Meshes

Gaussian Splatting (hereafter GS) is a recently emerged 3D visualization method. Unlike conventional point clouds or meshes, GS assigns to each of the countless points that comprise a scene a gently spread Gaussian distribution (a semi-transparent particle with bell-shaped shading) and layers these to depict surfaces. To use an analogy, it's like gradually layering countless small paint blotches on a canvas to create a picture that looks like a photograph. GS performs this in three-dimensional space, reproducing object shapes by smoothly interpolating between points. The result is the ability to generate extremely smooth, detailed, and photo-like 3D models.

So what sets GS apart from traditional methods? First, point clouds: point cloud data is a collection of surface coordinate points and, while measurement accuracy can be high, visualization often retains a grainy feel. Gaps between points or flicker (noise) depending on viewing distance and angle can make the model look coarse when moved in real time. Acquiring a higher point density improves detail but increases data volume and makes handling heavier.

Next, meshes (polygon models) construct object surfaces by connecting triangles based on point clouds. Because meshes represent continuous "surfaces," their appearance is smoother and they are suitable for integration with CAD drawings or BIM models and for analyses such as volume calculations. However, fine structures—such as scaffolding pipes or thin elements like wires—are prone to loss during mesh generation, and hole-filling or simplification can lead to shapes that differ from the real object. Also, making a mesh photorealistic by applying photographic textures requires advanced processing and, depending on model scale, can make the data very heavy and rendering time-consuming.

By contrast, Gaussian Splatting is an intermediate approach that uses the positional and color information of points like point clouds but represents points not by explicitly creating continuous faces like meshes, but as soft surfaces. Each point’s Gaussian-shaped splat overlaps and blurs boundaries, so the entire model appears as a continuous smooth surface. For example, a wall that looked rough in a coarse point cloud can appear as a seamless wall in GS because particles blend together. Individual splats can be adjusted in size and shape while retaining positional accuracy, so fine details can be reproduced faithfully to photographs. Complex scaffolding, piping, tree branches and leaves—elements hard to express with meshes—are easier to capture without loss in GS because it remains point-cloud-based.

Furthermore, GS’s rendering process is highly suitable for GPUs. Rather than ray-tracing through a neural network as conventional NeRF (Neural Radiance Fields) does, GS directly projects and composites each point (splat) per view, making rendering very fast. Even for large scenes containing millions of splats, with appropriate LOD (level of detail) management users can obtain smooth real-time imagery while moving the camera. In this way GS unites the accuracy of point clouds with the smooth appearance of meshes and, at the same time, enables interactive real-time operation—an innovative technology. Note, however, that models generated by splatting are primarily visualization-focused and are not suitable for precise CAD editing or geometric analysis (they lack explicit polygon boundaries and surface attributes). Bearing that in mind, GS is positioned as a digital twin representation for "making things look convincingly real."

The Value of Gaussian Splatting in Digital Twins

Using GS for on-site digital twins brings various benefits not available with conventional approaches. Key points include:

• Overwhelming site fidelity: You get sharp, high-definition 3D models that look as if you are viewing photographs. Because the site’s colors and textures are reproduced as they are, people can grasp conditions remotely as if they were on-site. Details such as fine cracks and sign text can be checked on the model, so GS’s ability to reproduce the site “as-is” in digital space is a major strength.

• Smooth, immersive display: GS models lack point-cloud flicker and don’t show polygon edges like meshes, so moving the viewpoint always yields smooth imagery. Walking through the model in a 3D viewer is less stressful and creates an immersive sense of being inside the space. This also works effectively in VR and AR applications.

• Fast rendering and lightweight data: Thanks to efficient GPU rendering, GS operates very smoothly. Whereas handling large point clouds or high-resolution meshes used to slow down display, GS can present and interact with large-scale site data at near-real-time speeds. Because you don’t need heavy mesh models or ultra-high-resolution textures, overall data processing is lighter, making cloud-based sharing and viewing smoother.

• Faithful representation of complex, fine structures: Thin structures such as scaffolding pipes, wires, and tree branches—elements often lost in conventional reconstruction—can be clearly modeled in GS. Since it is point-cloud-based, every measured point can become a visible element, enabling small components and complex shapes to be represented with minimal loss. This increases the completeness of the digital twin and makes detailed on-site review and record-keeping more effective.

• Intuitive information sharing: Realistic-looking digital twins are easy for non-experts to understand intuitively. For instance, site conditions that are hard to grasp from traditional point clouds or drawings can be instantly visualized in GS for all stakeholders. Reviewing a model during a construction meeting reduces misunderstandings and smooths consensus-building. The interactive nature—allowing inspection of details from needed viewpoints—also provides flexibility that photos or videos cannot.

Compatibility with LRTK and Benefits from Adoption

How the source 3D data is captured is important to fully leverage Gaussian Splatting. In that respect, the smartphone surveying solution LRTK is an excellent match for GS. LRTK uses a small RTK-GNSS receiver attached to a smartphone and a dedicated app to continuously position the smartphone at centimeter-level accuracy (half-inch accuracy) during photo capture and LiDAR scanning. Point clouds acquired this way come with accurate absolute coordinates (real-world coordinates) from the outset, and distortions during scanning are corrected in real time. As a result, the obtained point clouds and photos align with the site’s geodetic system without post-processing, and the high-precision positional information is directly reflected in models generated by GS. Photogrammetry-derived 3D models normally end up in arbitrary local coordinate systems, but by using LRTK you obtain digital twins that conform to the on-site coordinate system from the start.

Measurement with a smartphone + LRTK is an extremely easy and mobile way to collect data for GS. Without preparing a dedicated laser scanner or large-scale photographic rig, site personnel can gather high-precision point clouds and photos simply by walking the area with a smartphone in hand. Images captured are tagged with RTK-derived position data, so photogrammetry processing to generate point clouds or GS models later yields fast and stable results. Because the data are supported by accurate positioning, dimensions such as lengths and areas of structures are correctly reflected, so the finished GS models can be used not merely as pictures but as measurable digital twins. Point clouds scanned on different days also align precisely, so when generating GS models that span large areas, parts can be joined without positioning errors.

Moreover, LRTK has high affinity with AR, enabling the acquired 3D data to be overlaid on the real world. For example, if you overlay an LRTK-acquired, GS-processed as-built model on field footage in a tablet AR app, you can easily review past site conditions on the spot. You could also GS-model existing structures before construction and display them in AR alongside future BIM models to intuitively compare old and new. Because LRTK provides high-precision alignment, GS’s photorealistic models can be effectively used for such spatial augmentation. LRTK already includes AR functions to overlay design data and acquired point clouds for tasks like stakeout guidance and as-built verification, and combining those features with GS’s smooth as-built visuals makes on-site visualization even clearer and more persuasive.

In this way, LRTK is a key platform for applying cutting-edge technologies like Gaussian Splatting on-site. Adoption enables 3D measurement that balances "sufficient accuracy and overwhelming ease," dramatically improving the quality and applicability of resulting digital twins. Compared with dedicated equipment, initial costs can be greatly reduced, and the ease of use—even by non-specialists—is a major advantage. Workflows such as immediately uploading scanned data to the cloud to generate and share GS models become feasible. With LRTK, site staff can quickly scan necessary areas during routine construction management or inspections, convert them to digital twins, and instantly share and review the realistic models with stakeholders. This directly supports the rapid information sharing and decision-making that construction DX aims for.

Future Outlook for Construction DX: BIM/CIM Integration and Cloud Sharing Potential

Digital twins combining Gaussian Splatting and LRTK are expected to play a central role in promoting construction DX. In particular, integration with BIM/CIM will enable workflows that seamlessly handle planning and as-built conditions. For example, overlaying an LRTK-acquired GS model obtained during construction onto BIM data under design can quickly reveal deviations or construction errors. While point clouds are already imported into BIM software for clash checks and the like, GS models—with their rich visual information and intuitive clarity—should become even more useful references across design, construction, and maintenance phases. In the future, technologies may mature to efficiently derive CAD drawings from GS point cloud models or to refine GS representations using BIM models, further blurring the line between reality and design data.

In terms of cloud sharing and web display, GS broadens digital twin use cases. Hosting high-detail 3D models in the cloud and allowing stakeholders to freely change viewpoints via a web browser would enable remote site checks during online meetings. Large-scale 3D data that once required dedicated viewers or high-performance PCs can be streamed more easily with GS’s efficient compression and rendering methods, bringing mobile viewing within reach. Site supervisors checking high-fidelity models on a tablet and sharing instructions on the spot may soon become commonplace. Overseas projects are already appearing that build city-scale digital twins with GS and publish them on web platforms. In CG fields such as entertainment and VR training, photo-like 3D spaces will create new experiential value. In infrastructure inspection, disaster prevention, and urban planning, the value of sharing and analyzing realistic 3D spatial information via the cloud will only increase.

These technological trends align with government-led digital reforms. In Japan, i-Construction and CIM introduction guidelines recommend using 3D on site, and it is conceivable that photo-like realistic point cloud models could become new standard deliverables. Sharing digital twins in real time among stakeholders to enable rapid decision-making contributes to productivity improvements and work-style reforms in the construction industry. Gaussian Splatting is precisely the key technology to accelerate that realization. Coupled with the development of 5G and cloud computing, a world where remote stakeholders can instantly grasp site details is near. In the coming years, digital twins built with Gaussian Splatting are expected to become standard in the construction industry and an indispensable tool for DX promotion.

Conclusion

On-site digital twin technologies are evolving rapidly. Among them, Gaussian Splatting delivers immediately noticeable, overwhelming realism and represents an innovation that elevates the value of digital twins to the next level. The role of LRTK as the foundation for applying this on-site is also significant. A time is approaching when anyone can perform high-precision site scans with a smartphone and generate smooth, beautiful 3D models from that data.

"Why Gaussian Splatting now?" — the answer is that technical maturity and rising on-site demand intersect, making now the right time to adopt. GS combines expressive power and speed that were previously unattainable and has the potential to become a new standard. Moving early to embrace these cutting-edge techniques can streamline site work and differentiate you from competitors.

LRTK is a practical solution to take that first step. It brings advanced technologies simply to the field and makes state-of-the-art methods like GS usable at the site level. Why not take this opportunity to start experiencing digital twins with overwhelming realism through LRTK? Cutting-edge realism will bring new discoveries and value to your sites. As the boundary between reality and digital fades, let us together forge the future on-site scenes that this innovative technology will create.