Introduction

In recent years, technologies that turn terrain and structures into complete 3D data using drone aerial photography and laser scanners have been attracting attention in surveying and construction. Advanced 3D modeling, which used to be handled only by specialist surveyors or CG technicians, is becoming more accessible thanks to cloud processing and AI advances. However, there remain many concerns that prevent adoption: “expensive equipment and skilled operators are required,” “processing takes a long time even after capturing,” and “complex elements like vegetation and water surfaces are hard to reproduce.” In many sites, there has been a dilemma where “you can’t 3D-capture the location you need immediately” and “post-processing or sharing of obtained point cloud data takes time.” As a result, much of the collected point cloud data goes underutilized, and teams often revert to sharing information via drawings and photos as before.

A key solution to these challenges is the combination of smartphone-based high-precision positioning called LRTK and a new technique for rapidly generating high-fidelity 3D models from images called Gaussian Splatting. LRTK (a positioning system using a smartphone plus a small RTK-GNSS receiver) is an innovative measurement tool that attaches a pocket-sized device to a smartphone, enabling anyone to perform centimeter-level positioning easily. It allows accurate data capture with position information without special equipment or expert skills. On the other hand, Gaussian Splatting (GS), introduced in 2023, is the latest 3D generation method that can quickly reproduce photorealistic three-dimensional scenes from multiple images. This article explains the basic technology and features of Gaussian Splatting in plain language, explores its advantages through comparisons with conventional methods, and introduces how well it pairs with data acquired by LRTK and specific application scenarios to reveal the power next-generation 3D modeling brings to the field.

*Note: LRTK: positioning system using a smartphone + RTK-GNSS receiver (provided by Lefixea)*

Technology and Features of Gaussian Splatting

Gaussian Splatting (hereafter GS) is a cutting-edge technique for generating high-quality 3D models from multiple photographs. As the name implies, it uses the image of “splatting” — scattering countless tiny elements (Gaussian distributions) through space and projecting each as softly blurred colored particles to render surface textures smoothly. By assigning attributes such as color, opacity, and size to each element, GS reproduces textures that can be mistaken for photographs. Unlike traditional point clouds or mesh models, GS directly optimizes the entire scene from images to create a 3D representation, so it does not require intermediate point cloud generation or polygonization. As a result, details that used to produce noise—such as densely foliaged trees or reflections on glass and water—can be rendered as realistic 3D space.

Technically, GS draws inspiration from AI-based methods called NeRF (Neural Radiance Field) but is characterized by its use of explicit 3D primitives. While NeRF uses large computational resources to learn space’s radiance and density in a black-box manner, GS prepares numerous Gaussians (soft, spherical point representations) that compose the scene and adjusts their positions, shapes, and colors to match the photos. Because it does not allocate computation to empty regions of space, GS is efficient; recent research reports reducing training (model-generation) time to a few hundredths of that required by conventional NeRF, and rendering at near–real-time speeds.

In short, Gaussian Splatting is fundamentally different from “point clouds that simply display points” or “meshes that stitch surfaces together.” It is a new 3D representation that combines the lightness of point clouds with the expressive power of meshes. Numerous overlapping Gaussians softly cover object surfaces, producing smooth, gap-free models with reduced data size. GS is also well-suited to rendering optimizations on GPUs, and given only the captured images, it has the potential to construct and verify models on-site in a short time. These characteristics make GS notable as a next-generation 3D technology capable of pursuing both photographic realism and surveying-level accuracy.

Comparison with Conventional Techniques and Advantages of Gaussian Splatting

When creating 3D models, traditional approaches have included point clouds, meshes (polygons), and photogrammetry. Each is proven, but they have known limitations in terms of expressiveness and efficiency:

• Point cloud representation – Point cloud data acquired by laser scanners or photogrammetry is highly accurate, but because it is merely a collection of individual points, surfaces can appear sparse. Increasing point density makes surfaces smoother but leads to massive data volumes that are hard to handle. Point clouds also have limited color information by themselves, requiring additional processing such as projecting photos or converting to meshes with textures.

• Mesh representation – Generating polygonal meshes from point clouds or CAD data covers objects with surfaces and fills gaps, but achieving high-detail representation requires a huge number of polygons and results in bloated data. Structures with gaps or transparency—like tree leaves or wire fences—are difficult to reproduce with meshes, and forcing representation can demand enormous polygon counts or texture sizes. Materials that reflect or transmit light, such as water surfaces or mirrors, are also hard to express realistically with static meshes and textures.

• Photogrammetric representation – Photogrammetry using SfM/MVS to generate point clouds and meshes from drone photos is excellent at leveraging on-site imagery directly. However, it involves multi-stage processing and large computations, so generating final models takes time. Noise removal and filling missing areas often require manual adjustments and expert software skills. Achieving high accuracy can also increase on-site tasks, such as placing ground control points (GCPs), reducing responsiveness.

Against these weaknesses, Gaussian Splatting offers many advantages as a next-generation approach. As noted, GS describes scenes using many Gaussians with position, color, and opacity, enabling continuous representation from individual leaves to the sparkle on water. Sparse point cloud displays are filled because each point has a soft spread, and smooth surfaces emerge without a mesh. The modeling workflow itself proceeds end-to-end with image ingestion and optimization, significantly shortening processing time. In some cases, model generation can be completed immediately after on-site capture, allowing additional shooting or on-the-spot data checks as needed.

In terms of data size, GS models—built from collections of lightweight Gaussian primitives—tend to have smaller file sizes than conventional mesh models with comparable detail. Generated models can be visualized in dedicated viewers or converted to point data for export in conventional formats (LAS/PLY), enabling flexible integration with existing CAD and GIS software. Thus, Gaussian Splatting is expected to be effective across various sites due to its balanced strengths in expressiveness, efficiency, and data compatibility.

Compatibility of LRTK Data and Gaussian Splatting

To make the most of GS in the field, high-quality photographs and accurate position information are essential. This is where data acquired with LRTK proves powerful. With LRTK, anyone can easily collect photos with coordinate metadata and point cloud data. For example, attaching an LRTK receiver to a smartphone and walking a construction site while photographing will attach highly accurate, centimeter-level location coordinates to each photo. Traditionally, when creating 3D models via photogrammetry, workflows required matching common points across many images to estimate camera positions and placing target markers or measuring known lengths (ground control point placement) to align scales. Using photos whose positions are known via LRTK allows generation of 3D models already aligned to an absolute coordinate system, greatly reducing tedious alignment tasks.

Moreover, LRTK makes it easy to collect large volumes of photos and point cloud data on-site, which contributes to higher success rates and improved quality in GS model generation. Gaussian Splatting performs best when fed many images with high redundancy, so an LRTK-enabled environment that makes it easy to increase the number of captures is ideal. In addition to aerial drone captures, a hybrid data acquisition approach—where people on the ground with LRTK-equipped smartphones photograph details on foot—can cover hidden areas and forest interiors without omission. Because these data sets share RTK-based reference coordinates, drone and ground data can be combined later without misalignment and seamlessly merged into a single model.

Thus, the combination of LRTK, which allows anyone to easily collect accurately georeferenced data, and GS, which can automatically generate high-fidelity 3D models, represents a next-generation best match. On-site, non-expert operators can quickly measure and immediately 3D-capture and share results, dramatically shortening lead times from capture to model verification. For example, in disaster response, simply photographing with an LRTK-equipped smartphone can produce a photorealistic 3D model within minutes, enabling remote headquarters to share precise spatial information. The LRTK + GS combination’s accuracy, speed, and ease of use are likely to become a standard for future field measurement and modeling.

Specific Use Cases

• Construction sites: In infrastructure and building works, easy on-site surveys using LRTK combined with fast GS modeling allow accurate, realistic recording of construction status. For example, modeling excavation or embankment shapes in 3D and overlaying them with design data can quickly detect construction errors and streamline earthwork volume calculations. Tasks that previously required specialist surveying teams can be carried out quickly by on-site personnel using a smartphone, greatly accelerating schedule management and progress reporting. Sharing as-built models via the cloud enables remote offices to monitor construction status, improving safety management and client reporting.

• Disaster response: Rapidly 3D-capturing affected areas such as earthquake or landslide sites is crucial for recovery planning. Using LRTK-equipped drones for aerial coverage and workers on the ground photographing damaged structures with smartphones enables rapid generation of detailed integrated models. GS reproduces rubble piles and collapsed terrain directly from photos, allowing remote analysis of damage without entering hazardous zones. Teams can share models on the cloud with local governments and support organizations to speed damage assessment and restoration route planning. Volume measurement of debris and dimensional analysis of affected areas can also be performed accurately in 3D to aid recovery planning.

• Cultural heritage documentation: LRTK + GS is useful for preserving historic buildings and artworks. Compared to labor-intensive traditional 3D scanning, precise 3D archives can be created using only smartphone photography, lowering the barrier to cultural heritage surveys. GS can digitally record the texture of fine reliefs on sculptures and architectural ornamentation, supporting deterioration comparisons and restoration planning. Creating scale-accurate models facilitates replica production or VR exhibitions, providing a new way to pass valuable heritage on to future generations. Making precise measurement accessible without costly laser scanners or specialist contractors will encourage more comprehensive digital preservation of cultural assets.

• Exterior design: In home and garden remodeling, GS-generated as-built 3D models are powerful. Photographing house exteriors and site areas with an LRTK-compatible smartphone produces a dimensionally accurate “virtual site” quickly. Landscape designers and architects can use this base to test designs while viewing the real environment. GS accurately reproduces reflective materials like window glass and water features, making it easier to share completed-image concepts with clients and improving the quality of meetings and presentations. In real estate, exterior models can be published online to give remote customers realistic viewing experiences.

• Topographic surveying and environmental analysis: LRTK + GS is also effective for wide-area topographic surveys such as mountain areas and river basins. Areas hidden under trees or in cliff shadows in aerial photos can be supplemented by ground-level LRTK captures to ensure full coverage. High-resolution models can be used to generate contour lines and longitudinal sections, or to run simulations for earthwork volumes and flood extents—analyses traditionally performed on point cloud survey data. Because models are georeferenced, they can be overlaid on GIS maps for hazard analysis, or compared across time to monitor terrain changes. In disaster prevention, periodic 3D recordings of landslide-prone areas or shoreline erosion can detect subtle changes by model comparison, enabling monitoring applications.

Workflow and Key Points

• Data acquisition: Use GNSS-equipped drones or LRTK smartphones to capture many high-resolution photos of the target area. Planning for roughly 80% overlap between adjacent photos improves 3D reconstruction accuracy. For building exteriors, capture from both ground and air; for terrain, capture from multiple altitudes to eliminate blind spots.

• Model generation: Load the captured image data into dedicated software and the Gaussian Splatting algorithm will automatically generate a 3D model from the photo set. With a PC equipped with a high-performance GPU, processing hundreds of images can be completed in tens of minutes. Since intermediate steps such as point cloud generation and mesh conversion are unnecessary, model creation can be executed with a single button press.

• Use and sharing: The completed model can be intuitively viewed and measured in a dedicated viewer or web browser. Sharing a URL allows stakeholders to review the scene remotely. If needed, the model can be converted to point cloud formats (LAS/PLY) or mesh formats (OBJ, etc.) and imported into conventional CAD/GIS software for overlaying with design data, creating cross sections, and performing other analyses.

Conclusion

The combination of Gaussian Splatting and LRTK promises a major transformation in how 3D spaces are reproduced. High-precision 3D measurement that once required specialists is becoming achievable with familiar tools, and a future where anyone can handle photorealistic 3D models in minutes is within reach. This approach can reduce costs associated with outsourcing to survey companies and complex data processing, directly improving field productivity. Beyond complex site surveys, simple surveys and fixed-point monitoring will also become easier to conduct with 3D technology, dramatically improving operational efficiency and communication quality.

Indeed, in the ongoing DX (digital transformation) and digital twin trends in surveying and construction, the ease and accuracy of LRTK + GS are highly compatible. In Japan, GS rendering capabilities are already being integrated into civil engineering 3D software, and the infrastructure for practical use is progressing. Information that was difficult to convey with plans and photos alone can be shared three-dimensionally and intuitively using this system. If you face challenges in on-site 3D utilization, consider adopting a next-generation workflow that combines LRTK data capture with Gaussian Splatting model generation. You will likely experience unprecedented speed and quality in visualizing 3D spaces.

In the not-too-distant future, it may become commonplace for site personnel to casually scan their surroundings with smartphones and instantly share 3D models. LRTK and Gaussian Splatting will push the visualization of field sites into a new dimension.