Rayanix designs and builds 3D reconstruction systems, from camera calibration to textured, real-scale meshes, for industrial inspection, medical imaging, security, and retail analytics. If you are deciding how to capture a real object or space in 3D, this page explains the options and what to check before you trust the result.
What is 3D reconstruction?
3D reconstruction is the process of computing the three-dimensional shape of a real object or scene from images or scanner data. The result can be a point cloud, a surface mesh, or a textured model. The two main routes are image-based methods, which work from overlapping photographs, and scanner-based methods, which measure geometry directly. A 2023 peer-reviewed overview in Sensors covers both families and classifies scanners by operating principle [1].
Four ways to reconstruct, and what each is good for
| Method | How it works | Typically strong at | Watch out for |
|---|---|---|---|
| Photogrammetry (structure-from-motion + multi-view stereo) | Matches features across overlapping photos to recover camera poses and surfaces | Editable, scaled meshes from ordinary cameras [4] | Smooth, shiny or low-texture surfaces can produce holes or distorted regions [5] |
| Structured-light / laser scanners | Project a pattern or laser and measure its deformation | Direct geometry capture at close range; common in inspection | Reflective or translucent materials; hardware cost; needs calibration [1] |
| LiDAR | Measures distance by time-of-flight | Large spaces and fast capture | Resolution and accuracy vary widely by device class; in one 2026 building study, image-based methods outperformed phone LiDAR on visual and spatial metrics [3] |
| Neural radiance fields / Gaussian Splatting | Optimize a learned scene representation to reproduce views | Photorealistic, real-time viewing; the original Gaussian Splatting paper reports ≥30 fps at 1080p [2] | Designed for how a scene looks, not for measurement; spatial accuracy is still an active research question [3] |
No method wins everywhere. Many real projects combine them, for example photogrammetry for dependable geometry and splatting for appearance.
Which one do you actually need? Start with the question you must answer
- "Is this part within tolerance?" You need measurable geometry. Prioritize calibrated scanning or photogrammetry with a validation step against a reference.
- "Can a surgeon or customer see a realistic preview?" Appearance matters, but so does correct proportion, so scale and landmarks need to be controlled.
- "Can we explore this space remotely?" Visual fidelity may outweigh millimeter accuracy; splatting or textured meshes can fit.
- "Do we need to count, track or analyze?" Reconstruction is usually one stage in a larger computer-vision pipeline.
Why calibration and scale decide whether a model is usable
A model that looks right can still be wrong in size. Cameras have lens distortion and internal parameters that must be estimated; images alone do not fix the real-world scale unless a known reference is introduced [7]. A common practical remedy is placing known-size markers in the scene so the reconstruction can be scaled automatically [6]. Skipping calibration and validation is the most frequent reason a pretty 3D model fails in engineering use. [R]
What Rayanix builds
- Camera calibration and multi-camera systems for reliable geometry
- Textured 3D mesh reconstruction pipelines, including sparse-view capture of people and objects at real-world scale
- 3D model viewing and delivery in the browser or in the local application for common formats (
.glb,.gltf,.obj,.stl,.ply) - Industry applications across four areas: industrial quality inspection, medical and health imaging, security and surveillance, and retail and warehouse intelligence. See Industrial Quality Inspection.
[V]
How an engagement works
- Discovery: what decision will the 3D model support, and what accuracy does it need?
- Capture plan: hardware, lighting, markers and sample data.
- Reconstruction: build and tune the pipeline on your data.
- Validation: compare results against a reference measurement and report the error.
- Integration: deliver the model, viewer or API into your workflow.
Talk to an engineer
Tell us what you need to measure, inspect or visualize, and we will recommend an approach, even if the best answer is a method we don't build. Contact Rayanix
References
- Verykokou & Ioannidis, "An Overview on Image-Based and Scanner-Based 3D Modeling Technologies," Sensors 23(2):596, 2023. https://doi.org/10.3390/s23020596
- Kerbl et al., "3D Gaussian Splatting for Real-Time Radiance Field Rendering," ACM TOG 42(4), 2023. https://arxiv.org/abs/2308.04079
- "A Comprehensive Evaluation of the Spatial Accuracy of Building Gaussian Splatting," ISPRS Annals XI-M-1-2026.
- Varjo Teleport, "Photogrammetry vs NeRFs vs Gaussian splatting."
- KIRI Engine, "3DGS vs Photogrammetry vs LiDAR: A Practical Decision Guide."
- Plain Concepts, "Digital twins & 3D Gaussian Splatting."
- Hartley & Zisserman, "Multiple View Geometry"