RealityCapture for Heritage Photogrammetry: Complete Workflow

Why RealityCapture for Heritage

RealityCapture's defining characteristic is speed: it aligns and reconstructs large image sets dramatically faster than most competitors, often by an order of magnitude on big datasets, which matters when a temple complex generates thousands of images. Its second strength is mesh detail — it produces very high-resolution meshes that resolve fine carved stone extremely well. Its third is native handling of mixed inputs: it ingests laser scans alongside photographs and aligns them together in one project, which suits the hybrid photo-and-scan datasets common in heritage work.

The trade-offs are a steeper, less conventional interface, a component-based alignment model that behaves differently from Metashape's chunks, and historically a Windows-and-NVIDIA-GPU requirement. For heritage documentation specifically, RealityCapture is an excellent choice for detailed carved surfaces and large datasets where speed matters; the workflow below is oriented to getting rigorous, scaled, measurable output from it rather than just an attractive model.

Before You Start: Project Setup

Prepare images before importing, exactly as for any photogrammetry tool: process RAW files to a consistent white balance and exposure, and export high-quality images (16-bit TIFF or high-quality JPEG). RealityCapture reads image EXIF for initial camera parameters, so keep metadata intact. Organise images into a clear folder structure per subject, because a heritage project may combine several capture sessions.

Create a new project and add imagery through 'Inputs'. If the project includes laser scans, add them as inputs too — RealityCapture treats them as another data source to be aligned. Save the project early and often to a fast local drive; RealityCapture projects reference their inputs, so keep the imagery and project together and avoid moving folders mid-project.

• Batch-process RAW to consistent colour and exposure before import — inconsistent exposure degrades both alignment and texture quality.
• Keep EXIF metadata intact so RealityCapture can seed camera calibration from it.
• Add laser scans as inputs if you have them — RealityCapture aligns scans and photos together natively.
• Save to a fast local drive and keep inputs and project co-located; RealityCapture is I/O-intensive.

Alignment and Components

Press 'Align Images' to build the sparse reconstruction. RealityCapture finds features, matches them across images, and computes camera positions — very fast, but with an important conceptual difference from Metashape: the result is organised into 'components'. A component is a set of images that aligned together into one consistent reconstruction. Ideally you get a single component containing all images; if the software could not connect all images into one, you get multiple components — RealityCapture's equivalent of a split or partial alignment.

Multiple components are the first diagnostic to check after alignment. They mean some images did not link to the rest — usually because of a coverage gap, a featureless or blurred set of images, or two capture sessions with no shared views. The fixes are the same as any alignment failure: add connecting images that bridge the gap, add control points that appear in both components to tie them together, or remove bad images. Do not proceed to reconstruction with a fragmented set of components expecting one clean model.

Control Points and Scaling

This is the step that makes RealityCapture output measurable, and the step most often skipped. A freshly aligned RealityCapture model has arbitrary scale, position and orientation — it is geometrically correct in relative terms but has no real-world size. Before any measurement, orthophoto or export is valid, you must give it scale, and for georeferenced work, a coordinate system.

RealityCapture offers two mechanisms. Control points: place a control point on a feature or target, mark it in several images (RealityCapture's automatic projection helps propagate it), and either assign it real-world coordinates (from a total station or GNSS survey) for full georeferencing, or use control points in pairs to define known distances. Distance constraints: define the known real-world distance between two control points — even two well-measured scale bars give the whole model correct scale. For measured heritage documentation, survey-controlled points are rigorous; for smaller objects, calibrated scale bars with coded targets are efficient and accurate.

1. 1Add control points on distinct features or placed targets that are visible in multiple images.
2. 2Mark each control point in at least three to five images; use RealityCapture's projection assistance to speed this up and check the residuals as you go.
3. 3For georeferencing: enter the surveyed real-world coordinates of each control point and set the project coordinate system.
4. 4For scale only: define distance constraints between control point pairs using measured scale-bar or survey distances.
5. 5Include at least one extra control point or distance as an independent check — not used to define scale, only to verify it.
6. 6Review the control point errors: residuals should sit within your accuracy target; a high residual on one point usually means a marking error — re-check its projections.

Reconstruction: Building the Mesh

RealityCapture builds a mesh directly (its native output is a high-detail mesh, rather than a dense point cloud as the primary product, though point clouds can be exported). Choose the reconstruction detail: 'Normal Detail' is suitable for most heritage documentation and large sites; 'High Detail' resolves the finest carved ornament but is slower and produces very large meshes. Match the detail to the required output — High Detail on an entire temple complex generates an unmanageable mesh, while Normal Detail on a single carved panel may under-resolve the ornament.

Use the Reconstruction Region (the bounding box) to limit reconstruction to the heritage subject and exclude the surrounding ground, sky and background — this saves substantial time and produces a cleaner mesh. After reconstruction, the raw mesh will include some background, floating artefacts and noise; these are cleaned in the next stage.

Colorize, Texture and Clean

With the mesh built, generate the surface appearance. 'Colorize' projects per-vertex colour quickly for a first look; 'Texture' produces the high-quality texture maps needed for a heritage record, projecting the source photographs onto the mesh. For heritage documentation where colour fidelity matters — condition interpretation, material distinction — texturing from colour-managed source images (processed to a colour reference at capture) is what carries accurate colour into the final model.

Clean the mesh before or after texturing: use the selection and filtering tools to remove background, floating artefacts and disconnected fragments (the 'Filter Selection' and small-component removal tools handle most of this). As with all heritage cleaning, be conservative on the carved surface itself — remove obvious artefacts and background, but do not smooth away genuine surface detail. If the mesh needs simplification for downstream use, use 'Simplify' to a target triangle count, keeping a high-resolution master and deriving lighter versions for specific uses.

• Colorize for a fast preview; Texture for the deliverable-quality appearance.
• Texture from colour-managed source images so condition-relevant colour is faithful.
• Remove background and floating artefacts with selection/filter tools; keep a high-resolution master mesh.
• Simplify only derived copies for web or lightweight use — never the archival master.

Export: Mesh, Orthophoto and Point Cloud

RealityCapture exports the outputs heritage documentation needs. For the archival mesh, export OBJ (with material and texture files) or PLY — open, interoperable formats suitable for long-term retention. For measured elevation production, generate an orthographic projection (orthophoto) of the textured mesh along the facade plane, exported as a georeferenced or scaled image to trace in CAD. For integration with laser-scan and GIS workflows, export a point cloud as E57 or LAS/LAZ.

Whatever the output, confirm the export inherits the scale and coordinate system you established with control — an export from an unscaled or wrongly-scaled project carries that error into every downstream drawing. Record, with the deliverable, the control/scale reference used, the check-point residual, the reconstruction detail and the coordinate system. This processing metadata is part of a professional heritage deliverable.

RealityCapture vs Metashape for Heritage

Both are professional tools; the choice is about fit, not superiority. The table summarises the differences that matter for heritage practice.

Common Mistakes

• Exporting an unscaled model — a RealityCapture mesh with no control has arbitrary scale; scale and verify before any measurement or export.
• Ignoring multiple components after alignment — components mean the images did not all connect; resolve with bridging images or shared control before reconstruction.
• Using High Detail on an entire site — it generates unmanageable meshes; match detail to the output purpose and use the Reconstruction Region to limit scope.
• Skipping the check point — without an independent control point or distance not used in scaling, the model's accuracy is unverified.
• Texturing from inconsistent, non-colour-managed images — colour fidelity needed for condition interpretation is lost; process RAW to a colour reference first.
• Simplifying the archival master — keep the high-resolution mesh as the master and simplify only derived copies.
• Assuming Metashape habits transfer directly — the component model, scaling workflow and interface differ; the concepts transfer, the button-by-button steps do not.

Professional Practice

In professional practice, RealityCapture earns its place when speed and mesh detail are the binding constraints — large temple complexes, deeply carved surfaces, and projects combining laser scans with photography. Its speed changes what is feasible within a project timeline, letting a practitioner process in hours what would take a day elsewhere. That speed is a genuine advantage as long as it is not mistaken for a shortcut around rigour.

The professional discipline that must accompany RealityCapture is scale and accuracy control. Because the tool so readily produces a beautiful, convincing mesh from photos alone, the responsibility falls on the practitioner to insist on control points, distance constraints and an independent check before the output is treated as a survey. The deliverable is not the mesh; it is the mesh plus the record of how it was scaled and verified.

For heritage practices deciding between tools, the pragmatic position is that RealityCapture and Metashape are complementary. Understanding both — and being able to move a project between them — gives a practice the speed of one and the reporting rigour of the other, and avoids being locked to a single vendor's licensing terms, which for RealityCapture have proven changeable.