Photogrammetry for Heritage Conservation: Methods, Workflow, Software and Best Practices

What is Photogrammetry?

Photogrammetry is the science and technology of obtaining reliable geometric information from photographs. The word comes from the Greek roots for light (phos), writing (gramma) and measurement (metron) — literally, 'measurement by light writing.' It has been in use since the mid-nineteenth century, when it was first applied to make topographic maps from pairs of photographs taken from known positions.

The core principle is triangulation. If the same physical point is visible in two or more photographs taken from different positions, and if the position and orientation of those cameras at the moment of capture are known, the three-dimensional coordinates of that point can be computed by extending rays from each camera through the corresponding image point and finding where they intersect in space.

For most of photography's history, this triangulation required precisely known camera positions — which meant either survey instruments to measure them, or specially built survey cameras whose geometry was precisely characterised. The application of Structure from Motion algorithms, beginning in the early 2000s and becoming practically usable in software around 2010, removed that requirement. SfM can compute camera positions automatically from the photographs themselves, making photogrammetry accessible to anyone with a camera and the right software.

Structure from Motion Explained

Structure from Motion (SfM) is a computer vision technique that simultaneously recovers the three-dimensional structure of a scene and the positions and orientations of the cameras that photographed it, from a set of overlapping photographs taken from multiple viewpoints.

The process works in stages. In the first stage — feature detection — the algorithm examines each photograph and identifies distinctive local features (corners, edges, texture gradients) that can be reliably recognised across multiple images. The Scale-Invariant Feature Transform (SIFT) algorithm and its derivatives are the most widely used approaches; they identify features that are distinguishable even when the camera angle, scale and illumination change between photographs.

In the second stage — feature matching — the algorithm identifies which features in one photograph correspond to the same physical point in another photograph. This matching creates a set of correspondences between image pairs. From these correspondences, the relative positions and orientations of the camera pairs can be computed.

In the third stage — bundle adjustment — the algorithm takes all the pairwise camera position estimates and optimises them simultaneously, finding the consistent camera trajectory and scene structure that best explains all the observed feature matches across all photographs. The output is a sparse point cloud — a set of the matched feature points with their three-dimensional coordinates — and the estimated position, orientation and internal parameters of every camera.

The fourth stage — dense matching — uses the camera positions computed by SfM to generate a dense point cloud, in which nearly every pixel in every photograph contributes a measured point to the 3D model. This is computationally intensive and is where most of the processing time in a photogrammetry workflow is spent.

The output of dense matching can then be converted to a polygon mesh, and photographic texture can be applied to that mesh using the original photographs. The final textured mesh is a photorealistic 3D model of the surveyed subject at the accuracy level of the original photography and ground control.

Types of Heritage Photogrammetry

Heritage photogrammetry applications can be grouped into three main types, each with a different capture approach and appropriate accuracy level.

Close-range terrestrial photogrammetry is the most established type for detailed heritage recording. The camera is handheld or mounted on a tripod, and photographs are taken from the ground around and within the building or object. It is used for façade recording, interior documentation, detail capture of carved surfaces, sculptural recording, and small artefact documentation. Achievable accuracies range from 0.5 mm for small objects captured at close range, to 5 mm for building façades photographed from up to 20 m distance.

Drone photogrammetry (UAV photogrammetry) uses a camera-equipped drone to photograph the building or site from the air. It is used primarily for roof survey, upper storey documentation, site-wide topographic capture and orthophoto generation. Achievable accuracies with proper ground control are typically 2–5 cm for surveys at standard drone altitudes (30–60 m above ground). Drone photogrammetry cannot access interior spaces.

Underwater photogrammetry uses the same SfM principles to document submerged heritage — sunken ships, submerged archaeological sites, underwater architectural remains. The refraction of light at the water surface introduces systematic errors that must be corrected through the use of underwater-rated cameras with dome ports, and the processing requires additional correction steps not needed for above-water work.

Equipment: Cameras, Drones and Accessories

The camera is the most important piece of equipment in a photogrammetry workflow, and also the most flexible. SfM processing can work with images from a smartphone camera; the quality of the results simply scales with the quality of the input imagery. For professional heritage documentation, where accuracy and detail matter, a high-resolution mirrorless or DSLR camera with a prime lens is the standard choice.

Key camera specifications for heritage photogrammetry are: sensor resolution (a minimum of 24 megapixels for detailed façade work; 36–50 MP for high-detail capture); a prime lens of 24–50 mm equivalent focal length (prime lenses have more consistent geometry than zooms, which matters for the geometric accuracy of the model); manual exposure control; and RAW file capability (RAW images retain more colour information for texture quality than compressed JPEGs).

For drone photogrammetry, the integrated camera systems on purpose-built survey drones — the DJI Phantom 4 RTK, the WingtraOne, the senseFly eBee — are calibrated and optimised for photogrammetric processing. Consumer drones such as the DJI Mavic 3 Enterprise produce good results for heritage surveys that do not require centimetric accuracy, and their portability is an advantage on sites with limited access.

Accessories that are not optional for professional heritage photogrammetry include: Ground Control Point (GCP) targets — marked panels placed in the scene and measured to known coordinates with a total station or RTK GNSS — and a colour reference chart for consistent colour calibration across image sequences captured under varying light conditions.

Data Capture Workflow

Good photogrammetry results depend more on disciplined capture planning than on any processing technique. The algorithm has strict requirements for image overlap; meeting those requirements consistently on a complex heritage building requires a systematic approach.

The standard terrestrial capture pattern for a building façade is a series of horizontal strips, with the camera at each position pointing perpendicular to the wall surface. Within each strip, the camera moves laterally with sufficient overlap — at least 60%, typically 70–80% — between adjacent positions. Multiple strips at different heights cover the full elevation. Crucially, overlap between strips is as important as overlap within strips.

For complex architectural elements — sculptural panels, column capitals, corbelled eaves — the standard strip pattern is not sufficient. These require an orbital capture sequence, with the camera moved around the element capturing images from many angles, including oblique angles that reveal undercuts and concealed faces. Deep recesses may require the camera to be introduced into them on a pole or fixed mount to capture their inner surfaces.

For drone survey, the standard mission type is a grid flight at a consistent altitude, with perpendicular cross-strips added for complex roofs. Oblique capture — with the camera tilted 45° or more from vertical — is increasingly standard for heritage surveys, as it provides elevation data that nadir (straight-down) photography misses.

Image exposure should be consistent throughout a capture sequence. Variable exposure across a sequence produces colour inconsistencies in the final texture that are visually distracting and make interpretation of surface condition less reliable. Manual exposure settings — consistent ISO, aperture and shutter speed across each capture session — produce better textured models than automatic exposure. A calibrated grey card image at the beginning and end of each session allows consistent colour correction in post-processing.

Ground Control Points (GCPs)

Ground Control Points (GCPs) are marked targets placed within the scene and measured to known three-dimensional coordinates using a total station or RTK GNSS system. They are introduced into the photogrammetry processing to tie the computed model to a real-world coordinate system, and to provide an independent check on the geometric accuracy of the reconstruction.

Without GCPs, a photogrammetry model is 'floating' — its internal geometry may be accurate, but its scale and position are not defined relative to any external reference. For most heritage documentation purposes, this is not acceptable. The model needs to be aligned with survey control data, and measurements taken from the model need to be verifiable against independently measured dimensions.

GCPs should be distributed across the scene — not clustered in one area — and should include some at the top and some at the base of the structure if the façade extends over a significant height. The minimum practical number is five for a simple façade; for a complex building, ten to twenty GCPs distributed around the structure and at different heights is more appropriate.

Check points — a further set of measured points that are not used in the processing but only in the accuracy assessment — are the standard way to validate a photogrammetry model's geometric accuracy. If the model's positions at the check points agree with the total station measurements to within the expected accuracy tolerance (typically ±5 mm for heritage work), the model is validated.

Processing Software Comparison

The photogrammetry software market for heritage applications is dominated by a small number of platforms. They differ in accuracy, speed, workflow flexibility, price and the learning curve they impose on new users.

Agisoft Metashape (formerly PhotoScan) is the most widely used professional photogrammetry tool for heritage documentation globally. It offers a mature, reliable processing pipeline with extensive control over processing parameters at each stage, supports a wide range of camera types and models, and produces results of high geometric and textural quality. The professional licence costs approximately USD 3,499; a standard licence (without some advanced features) is USD 179. Processing speed on a standard workstation is adequate for most heritage projects; GPU acceleration is supported and significantly reduces processing time.

RealityCapture by Capturing Reality (now owned by Epic Games) is the fastest commercial photogrammetry processing tool available, typically three to five times faster than Metashape on equivalent hardware, with comparable or in some workflows superior geometric accuracy. Its pricing moved in 2023 to a per-input-credit model under Epic's free tier for projects below a revenue threshold, which makes it cost-effective for many heritage users. The interface is less intuitive than Metashape for users new to photogrammetry.

Pix4D Mapper is widely used for drone mapping and survey, with strong integration with the DJI ecosystem and a cloud processing option that reduces the need for a high-specification local workstation. It is less commonly used for close-range heritage work, where Metashape and RealityCapture have stronger positions.

OpenDroneMap (ODM) is the primary open-source option, freely available and producing good results for drone surveys. It lacks some of the fine-grained processing controls of the commercial tools and is typically slower, but for heritage projects with budget constraints it provides a viable processing path.

For Indian heritage projects where cost is a significant constraint, the practical recommendation is Metashape Standard for terrestrial work (affordable, reliable, well-documented) and OpenDroneMap for drone survey processing (free, adequate for most heritage purposes).

Outputs and Their Uses

A photogrammetry processing pipeline produces several output types, and understanding what each is useful for helps in deciding which to produce for a given project.

The dense point cloud is the primary geometric output — a large set of points with x, y, z coordinates and RGB colour values, representing the surface of the surveyed subject at the density of the photography. Dense point clouds for close-range façade surveys of a major temple elevation might contain 50–200 million points; drone surveys of a large site might produce billions. Point clouds in LAS/LAZ or E57 format can be delivered to structural engineers, conservation architects and GIS professionals as an interoperable product.

The polygon mesh is a connected surface derived from the point cloud by triangulating between adjacent points. Meshes are more computationally tractable than point clouds for visualisation and 3D printing, and they can carry photographic texture applied from the source images. High-quality textured meshes are the standard output for public presentation, virtual tours, documentation archives and HBIM model generation.

Orthophotos are planimetrically corrected photographic images derived from the mesh — essentially aerial or façade photographs in which the effects of perspective have been removed, so that distances can be measured directly on the image. Orthophotos are the standard output for condition mapping: surveyors draw condition annotations (cracks, weathering zones, repair areas) directly on the orthophoto in a GIS or CAD environment, creating a scaled condition plan that can be updated at each future survey.

Digital Surface Models (DSMs) from drone photogrammetry give terrain elevation data for archaeological sites and historic landscapes, enabling the identification of earthwork features, buried structures and topographic analysis.

Accuracy and Quality Control

The accuracy of a photogrammetric survey depends on the resolution and quality of the input photographs, the consistency of image overlap, the number and distribution of ground control points, and the processing parameters used. For heritage documentation purposes, a useful rule of thumb is that geometric accuracy will typically be two to three times the size of the ground sampling distance (GSD) — the distance on the ground represented by a single image pixel.

For a camera with a 36 MP full-frame sensor and a 35 mm lens, shooting a building façade from 10 m distance, the GSD is approximately 2 mm. Expected geometric accuracy would then be 4–6 mm, which is within the standard tolerance for heritage drawings at 1:50 scale (where 1 mm on the drawing represents 50 mm on the building — so an accuracy of 5 mm is represented by 0.1 mm on the drawing, below the plotting resolution).

Quality control in photogrammetry processing should happen at multiple stages, not only at the end. After the sparse point cloud and camera alignment stage, check that camera positions are plausible — they should form a spatial pattern consistent with the capture sequence, with no wildly misplaced cameras. After ground control integration, check residuals at each GCP: residuals larger than the expected accuracy suggest either a measurement error in the GCP, a problem with the point identification in the images, or a processing error. Check point errors (from independently measured points not used in the processing) are the standard final accuracy verification.

Photogrammetry vs Laser Scanning

The question of whether to use photogrammetry or terrestrial laser scanning for a heritage project comes up on almost every significant documentation commission. The honest answer is that they are complementary rather than competing methods, and the choice depends on the specifics of the project rather than on a general superiority of either approach.

Laser scanning has advantages in speed for large interior spaces (a single scanner position captures millions of points in a few minutes without requiring any camera movement planning), in consistency of point density (the scan pattern is regular and predictable, unlike photography whose overlap can vary), and in the capture of complex concave geometries where photogrammetry can struggle with surface orientation issues.

Photogrammetry has advantages in cost (a professional camera system costs a fraction of a professional laser scanner), in the simultaneous capture of colour and texture alongside geometry (laser scanners typically capture colour at lower resolution, from a separate camera run), in the ability to access geometry that would be dangerous or impractical for a laser scanner setup (high above the floor, in confined spaces accessible by pole), and in the quality of surface texture on coloured or polychrome surfaces.

For most heritage projects that can afford only one method, photogrammetry is the more accessible choice and adequate for the majority of documentation purposes. For high-value projects where completeness and speed of capture matter, and where the budget allows, combining laser scanning for interior geometry with photogrammetry for surface texture and exterior detail produces the most comprehensive record.

Photogrammetry for Heritage in India

Photogrammetry has been applied to Indian heritage documentation for two decades, with the pace of adoption accelerating significantly since drone-based survey became technically and economically accessible around 2015–2018.

The ASI has used photogrammetry in heritage documentation projects at major centrally protected monuments including the Sanchi Stupa, the cave temples of Ellora and the temple complexes of Tamil Nadu. Academic institutions — particularly those with architecture or heritage conservation programmes — have used it for research documentation. Private consultancies and heritage conservation firms use it on commissioned projects for temple trusts, private heritage owners and government agencies.

The particular challenges of photogrammetry for South Indian heritage deserve specific mention. Dravidian temple architecture presents several difficulties that are not typical of the building types for which most photogrammetry guidance has been written: the heavily carved gopuram towers, often 40–60 m tall, require both close-range capture for detail and drone capture for upper portions; the dense sculptural programmes on wall surfaces require high-resolution close-range photography from scaffolding or elevated positions; the enclosure of major temple complexes within prakaram walls creates interior environments with variable and often challenging lighting; and the active religious use of most significant temples creates access and timing constraints on survey operations.

Drone photogrammetry of Indian temples also requires engagement with ASI and, in many cases, with the managing Devasthanam or Hindu Religious and Charitable Endowments (HR&CE) department. The DGCA's Digital Sky platform has simplified the permissions process for commercial drone operations, but the multi-party permission landscape for a major religious site can still require several months of preparation. Early engagement with all relevant authorities is essential to realistic project planning.

The growing availability of photogrammetry processing capacity — particularly through cloud-based processing services — is gradually reducing the hardware barrier. Projects that previously required a dedicated high-specification workstation can now be processed through cloud services at project cost, making photogrammetry documentation accessible to a wider range of heritage owners and practitioners across India.

Common Mistakes in Heritage Photogrammetry

The most common error is insufficient image overlap. The SfM algorithm requires substantial redundancy in the photographic coverage to compute accurate camera positions; an overlap of 70–80% between adjacent images is a standard starting point, not a maximum. Areas with less overlap — the edges of a façade, the top of a wall, areas obscured by vegetation — are where reconstruction failures most often occur. These gaps are extremely difficult to fix in processing and almost always require a return to site.

Uniform, featureless surfaces — renders, plain plasterwork, uniform stone — challenge SfM matching because there are few distinctive features to track between images. In these situations, the algorithm produces a degraded point cloud or fails entirely. For genuinely featureless surfaces, projected light patterns (dot or stripe patterns from a projector) can create temporary texture that the algorithm can match. Alternatively, physical targets placed on the surface can provide artificial features. Failing to plan for this problem before arriving at a featureless-walled site is a common cause of project failure.

Poor lighting conditions significantly affect photogrammetry quality. Strong direct sunlight creates high-contrast shadows that obscure detail, cause exposure inconsistencies across the image sequence, and confuse the matching algorithm. Overcast conditions are generally preferred for photogrammetry capture; if direct sunlight is unavoidable, consistent lighting direction across the sequence (not mixed sun and shade within a single capture set) is the priority.

Moving objects in the scene — people, vehicles, shadows of moving tree branches — create features that appear in some images and not others, which confuses the matching algorithm and can cause artefacts or errors in the reconstruction. Scheduling photogrammetry capture at times of low activity, or explicitly cleaning up identified moving object masks in processing, reduces this problem.

Finally, failing to document the survey process alongside the data is a consistent omission. A photogrammetry dataset without notes on the camera used, the flight parameters, the GCP measurement method, the processing software and settings, and the accuracy verification results is of substantially lower long-term value than a fully documented dataset. The documentation of the documentation is not an afterthought — it is part of the work.