An orthomosaic is what a satellite image would look like if you could fly the satellite three feet over your job site. It’s the most-shipped drone deliverable in the world, and the math that turns a stack of obliquely-rotated photographs into one true- scale map is older than the drone industry by a century.
§ 01Drag the slider
Left side: a chaotic mosaic of raw drone photographs, each shot from a different angle. Right side: the orthomosaic that the photogrammetry pipeline produces — every pixel rectified to a top-down, true-scale projection.
ORTHOMOSAIC
RAW IMAGES
‹›
FIG · Raw oblique photos (left) reprojected to a north-up orthomosaic (right) at true scale.
ORTHOMOSAIC
A geometrically-corrected aerial image with uniform scale, suitable for measuring distances and areas directly. Produced by stitching many overlapping photographs and reprojecting them onto a digital elevation model.
§ 02The Structure-from-Motion pipeline
Every photogrammetry tool — open-source ODM, commercial Pix4D, Reality Capture, and the rest — implements the same five-stage pipeline:
1 · Feature detection
Find distinctive corner-like features in every image (SIFT, AKAZE).
2 · Matching
For each pair of images, find features that appear in both (FLANN).
3 · Sparse reconstruction
Triangulate camera poses and a sparse point cloud (RANSAC + bundle adjustment).
4 · Densification
Fill in millions more 3D points using multi-view stereo (PMVS / MVS).
Stage 3 — bundle adjustment — is where photogrammetry stops being “match features” and starts being “solve a global system of equations.” Every camera pose, every 3D point, and every sensor calibration parameter is optimized simultaneously to minimize reprojection error across thousands of constraints:
§ 04DSM, DTM, and what each is good for
Two different kinds of elevation model come out of the same pipeline:
DSM (Digital Surface Model)
Includes everything Sea level → top of canopy / roof. Use for: building heights, treetop volumes, line-of-sight analysis.
DTM (Digital Terrain Model)
Bare earth, with vegetation and buildings filtered out. Use for: drainage, contour lines, earthworks volume.
nDSM (Normalized DSM)
DSM minus DTM = object heights only. Use for: building footprints, vegetation height, change detection.
Hillshade
A shaded-relief render of the DSM. Use for: visualizing terrain in a single image.
§ 05File formats — what to ship
GeoTIFF (.tif)
The universal raster format. Every GIS reads it. Embeds CRS in the file.
Cloud Optimized GeoTIFF (.tif)
Same format, organised so HTTP range-requests can stream tiles. The web-friendly variant.
LAS / LAZ
Point cloud format. LAS is uncompressed, LAZ is the compressed variant (~10× smaller).
OBJ / glTF / GLB
3D mesh formats. glTF is the modern web-friendly choice; OBJ is the legacy fallback.
PMTiles
Static-tile archive — single file, no server, scales to terabytes. Great for shareable maps.
§ 06Open vs commercial pipelines
The open-source ODM stack — OpenDroneMap, NodeODM, WebODM — implements the same pipeline as commercial tools at no licence cost. Quality is competitive on most jobs and the math is identical (everyone ultimately calls into the same handful of computer vision libraries). Pix4D and Reality Capture are faster on very large datasets and have more polished UIs, but for solo operators and small firms, ODM is what most professional deliverables are processed through.