What is an orthomosaic? How a drone orthomosaic map is made, checked and flown

Orthomosaic, orthophoto and aerial photo

An aerial photo is a central projection: everything is seen from one point, so building tops lean away from the centre, hills come out bigger than valleys, and a tilted camera stretches one side of the frame. You can look at it, but you cannot measure on it.

An orthophoto removes those effects. The USGS describes orthoimagery as combining the image of an aerial photograph with the geometry of a map: a uniform-scale image corrected for feature displacement, using ground control, the camera's calibration and an elevation model. The correction is called orthorectification.

An orthomosaic is an orthophoto built from many photos at once, blended along seams into one image. With drones the words are used for the same file: ODM calls it an orthophoto, Pix4D and DroneDeploy an orthomosaic. A satellite or aircraft orthophoto of a whole county is often a single frame per tile; a drone site needs hundreds.

How a drone orthomosaic is made

Overlap first. Each point on the ground must appear in several photos, so the software can find the same feature from different angles. Front overlap is between photos along a flight line, side overlap between neighbouring lines.

Structure from motion (SfM) then matches thousands of features across the photos and solves for each camera's position and orientation, plus a sparse cloud of points. Dense matching fills that in to a dense point cloud, and from it the software builds a digital surface model (DSM): the height of the top surface, roofs and trees included. A terrain model (DTM) is the bare ground with those removed.

Orthorectification projects every photo straight down onto the DSM, so each pixel lands at its map position, and the mosaic step picks the best photo for each area and blends the seams. Where the DSM is rough, at roof edges, trees and wires, the image can smear or show a doubled edge: that is the surface model, not the camera.

Minimum front and side overlap from Pix4D's image acquisition guide, and the default Skyplanr plans with. Open water with no land in the frame cannot be reconstructed.

How to fly for an orthomosaic

A plan for a field or a building site with a DJI Fly drone, from the resolution to the processing.

1. 01

   Pick the GSD the job needs

   Start from the smallest thing the map must show: a kerb, a crack, a survey mark. A pixel several times smaller than it is the target. 2 cm/px is a common choice for a building site.

2. 02

   Work out the height

   GSD grows with height. With an Air 3S in the 12 MP mode, 2 cm/px is about 57 m above the ground; the mapping guide has the table for every DJI Fly drone. On sloping ground the pixels get bigger where the ground drops away.

3. 03

   Set the overlap for the terrain

   80% front and 70% side is a sound default for buildings and mixed ground. At 57 m with an Air 3S, that is a photo about every 12 m and lines about 25 m apart. Raise it over forest, crops, snow and sand, where features repeat or move.

4. 04

   Fly a straight-down grid

   Camera at -90°, lines along the long side of the area, and one or two lines beyond the edge so the border has the same overlap as the middle. Photos taken while the drone hovers or flies slowly enough to avoid motion blur.

5. 05

   Keep the light and the exposure steady

   Fly in even light, ideally overcast or with the sun high, and lock the exposure and white balance where the app allows. Moving cloud shadows and a sun low enough to throw long shadows make seams and gaps in the mosaic.

6. 06

   Add ground control and checkpoints if position matters

   For a map that must line up with survey data, mark and measure ground control points with an RTK receiver or a total station, and keep at least 20 more as checkpoints that the processing never sees. For a visual map of a site, relative accuracy is often enough.

7. 07

   Process and read the report

   Check the share of photos that aligned, the achieved GSD against the planned one, and the coverage of the area. The worked example shows a real report set against its requirements.

Accuracy: relative and absolute

Relative accuracy is how well distances and shapes inside the map agree with each other. A drone orthomosaic built from good overlap is usually consistent to a few pixels, so a measured length or area can be close even when the whole map sits off.

Absolute accuracy is how far the map is from true coordinates. A drone's ordinary GNSS puts the whole mosaic about 1-3 m off horizontally, and height can be off by much more. Ground control points (GCPs) or an RTK drone tie it to coordinates; checkpoints, measured the same way but left out of the processing, show how far off the result is.

The US National Standard for Spatial Data Accuracy (NSSDA) asks for at least 20 checkpoints spread over the area, from an independent source of higher accuracy, and reports horizontal accuracy at 95% confidence as 1.7308 × RMSEr when the x and y errors are similar. The ASPRS standard, Edition 2, asks for 30 checkpoints and reports RMSE. Quote accuracy with the test behind it, never the GSD.

File formats: GeoTIFF and COG

An orthomosaic is almost always delivered as a GeoTIFF: a TIFF image with the coordinate system and the position of each pixel stored inside it, standardized by the OGC. QGIS, ArcGIS, AutoCAD and most survey software open it directly. OpenDroneMap writes it as odm_orthophoto.tif, with the DSM and DTM as GeoTIFFs next to it.

A Cloud Optimized GeoTIFF (COG) is a regular GeoTIFF organized in internal tiles and lower-resolution overviews, so a web viewer can fetch only the part on screen with HTTP range requests. The OGC published it as a standard in 2023. A desktop GIS opens a COG like any GeoTIFF.

• GeoTIFF or COG: the map image with its coordinates, for GIS and CAD.
• DSM and DTM as 32-bit GeoTIFFs: heights for volumes, profiles and contours.
• KMZ or map tiles: a lighter copy for Google Earth or a web map, not for measuring.
• A quality report: photos used, GSD, overlap and, with GCPs, the residuals.

Questions

Related guides

• Drone mapping with DJI Fly: plan the grid, then build the orthophoto

  Drone surveying and mapping with the Mini 4 Pro, Air 3S and other DJI Fly drones: the grid as a KMZ, GSD per height, 80/70% overlap, flights and accuracy.

• LoD vs GSD: what an LoD3 request asks for, and how to fly it

  LoD is a 3D model's geometric detail, GSD is the photo's pixel size, and neither is accuracy. How to turn an LoD2 or LoD3 request into a DJI Fly plan.

• A drone mapping job, requirement by requirement: the plan, the ODM report and what is unverified

  One drone survey job as requirements versus results: what was asked, what was delivered, the evidence, and which results are met, not met or unverified.

Sources

Skyplanr is not affiliated with the third-party guides linked here.

1. USGS: High Resolution Orthoimagery (data.gov metadata, 2019): uniform scale, corrected for displacement, from ground control, sensor calibration and a DEM (catalog.data.gov)
2. Pix4D: selecting the image acquisition plan type (overlap for the general case, fields, forest, snow and sand) (support.pix4d.com)
3. FGDC: National Standard for Spatial Data Accuracy (FGDC-STD-007.3-1998), 20 checkpoints, 95 % confidence (fgdc.gov)
4. LIDAR Magazine: The ASPRS Positional Accuracy Standards, Edition 2 (30 checkpoints, RMSE) (lidarmag.com)
5. OGC: GeoTIFF Standard 1.1 (OGC 19-008r4) (docs.ogc.org)
6. OGC: Cloud Optimized GeoTIFF Standard 1.0 (OGC 21-026, 2023) (docs.ogc.org)
7. OpenDroneMap documentation: outputs (GeoTIFF orthophoto, DSM and DTM) (docs.opendronemap.org)
8. DJI Developer: DJI WPML overview (the KMZ format DJI Fly reads) (developer.dji.com)
9. DJI: Air 3S specifications (camera, flight time, weight) (dji.com)