Orthophoto from Drone Photos: A Photogrammetry Guide

A flight's photos are not a product on their own; to become measurable they have to be tied together and placed on the map through photogrammetry. CartaX does this in a single flow: you point it at the photo folder, the outputs you selected are produced together, and they are uploaded straight into your CartaX project. You run production either in the GPU-accelerated CartaX cloud (from the Photogrammetry tab of the upload dialog in the web app, or with Studio's Produce in the cloud button) or on your own machine with CartaX Studio. This guide explains what to watch for to get the best result. For installing Studio see What is CartaX Studio?.

What is produced?

In the same run, as many as you tick:

Output	Format	What it is for
Orthophoto	GeoTIFF (COG)	A scaled, georeferenced aerial image; the base of the map. Produced in every run.
DEM (DSM/DTM)	GeoTIFF (COG)	Surface and terrain elevation models; elevation readout, slope, volume and cut/fill on the map are computed from these. Selected together with the orthophoto.
Point cloud	LAZ → Octree	A dense 3D point cloud; measurement, cross-sections, AI classification (ground / vegetation / building / road / vehicle).
Textured 3D model	3D Tiles 1.1	A textured 3D model on the map; presentation and visual inspection.

Selecting only orthophoto and DEM shortens the run noticeably; the 3D model and the dense point cloud are the heaviest steps — in the cloud the dense point cloud is computed on a GPU.

Preparing the photos

Result quality comes from the photos more than from the engine:

• Overlap: at least 75% forward and 65% side. Low overlap = holes in the model, drifting edges.
• Photo count: hundreds of photos are normal; both Studio and the web upload dialog scan the root folder including subfolders (JPG/PNG/TIF) and find RTK .MRK files and side files by themselves.
• Positioning: photos must carry EXIF GPS. On RTK drones (e.g. DJI .MRK) centimetre-level positions are evaluated automatically.
• Avoid: moving water surfaces, uniform snow/sand, very dark shadows, mixing different lighting conditions in one run.

Positional accuracy: EXIF, RTK, GCPs

Studio handles three sources together:

• Position source: Automatic (RTK when .MRK is present, otherwise EXIF), EXIF only, or RTK required (the job does not start without .MRK — you never lose the accuracy you expected silently).
• A ground control point (GCP) file is optional. Plain text, one point per line: name E N Z or index name E N Z; lines starting with # are comments. The coordinate system (EPSG) is detected from the file, or you enter it by hand (e.g. 5254 TUREF/TM30).
• When a file is added there is one question: what should these points be used for?
  • Report only — the points do not enter production; after the run the deviation is measured and written to the accuracy report.
  • Find automatically — targets are detected in the photos and enter the solution (requires the RTK position file); undetected ones stay for review.
  • Mark manually — you mark the points on the photos yourself on screen; RTK is not required.
• Without a file, Enter and mark points lets you type coordinates and mark them on the same screen.
• You can keep a point out of the solution: it then becomes an independent check point and gives a real accuracy figure in the report.

The deviation of a point that entered the solution is not an independent measure. To prove accuracy, leave at least one or two points out of the solution.

Production quality

Production runs at a single quality: High — the highest-resolution point cloud and orthophoto, and a 3D model with crisp building edges and fully textured facades. No lower tier is offered, so measurements and imagery stay consistent; to shorten the run, reduce the number of outputs (e.g. drop the 3D model or classification).

Delivery coordinate system

The engine's raw output is always UTM. You choose the delivery:

• Automatic (recommended) — with a GCP file the products are delivered in its coordinate system, otherwise in UTM.
• Keep UTM — no transformation.
• Different EPSG — enter a code for delivery (e.g. 5254).

Under advanced settings there is Fit products to my points (datum correction): the constant offset between the GCPs and the products is measured and applied to the orthophoto/DEM/point cloud — no re-production needed (the mesh is not affected). If the deviation is still large after correction the problem is not position but the model's internal geometry; include the points in the solution and re-produce.

Where should it run: this computer or the cloud?

• Produce on this computer (Studio) — no tokens are spent and the raw photos never leave your machine. On the first production the engines download once (~0.4 GB). Studio uses all cores; with an NVIDIA GPU, dense point-cloud reconstruction moves to the GPU. The wizard measures free disk and memory beforehand; if they are insufficient the local button is disabled.
• Produce in the cloud (Studio, or the Photogrammetry tab on the web) — the photos are uploaded, production runs on CartaX's NVIDIA GPU cloud machines, and the result lands in your project. Billed in tokens: the charge follows the total megapixels processed and the number of outputs does not change the price; as soon as the photo count is known the estimated tokens and your balance are shown, and you are warned before the job starts if tokens are short. Up to 2,000 photos per job. Once the photos are uploaded you can shut the computer down.

Time expectation: a few hundred 45 MP photos → a few hours in the cloud; locally longer depending on hardware, and a machine with an NVIDIA GPU shortens the run noticeably. Several jobs can be queued; processing runs one at a time, uploads run in parallel. A running job can be suspended and continued with Pause / Resume; even if Studio closes, a local job resumes from the stage it reached at the next launch. If the upload fails, Retry upload repeats only the upload.

Where does the result go?

• On your computer: product folders under Documents / CartaX Studio / <flight name> — Ortofoto (orthophoto), SYM (DSM/DTM), Nokta Bulutu (point cloud), Mesh, Rapor (report). Rapor holds the engine report (report.pdf), the accuracy summary (accuracy.json), camera positions (shots.geojson) and the run logs.
• In CartaX: when the job finishes the outputs are uploaded to the project you chose. The orthophoto becomes a map layer; if you selected DEM, the layer panel offers Measurement elevation: Surface / DSM / DTM; point cloud and mesh open in their own viewers. While uploading, the CartaX Studio card on the web dashboard shows progress.

Common questions

Orthophoto edges are blurry or warped. Overlap drops at the edges; extend the flight one strip beyond the area of interest.

GCPs were not found automatically. Automatic detection needs the RTK position file and expects targets to be clearly visible in the photos; mark undetected points by hand on the same screen.

There is a constant offset between the orthophoto and cadastre/design data. With EXIF-only positioning a metre-level offset is normal; RTK or GCPs included in the solution close it. Make sure the delivery EPSG matches your project.

It takes very long. Drop the 3D model and classification; run on a machine with an NVIDIA GPU or send the job to the GPU-accelerated cloud. The run log is in the Rapor folder.

Related guides

• What is CartaX Studio?
• DSM vs DTM: what is the difference?
• Volume and cut/fill calculation
• Viewing point clouds in the browser