Technical15 min read18 February 2026Examin Engineering

Photogrammetry to ±2 cm: mission planning that holds up

A volume figure is only as good as the flight that produced it. When a reservoir capacity or a stockpile quantity ends up in a dispute, these are the parameters that decide whether your number survives scrutiny.

Ground sample distance sets your ceiling

GSD is the ground distance covered by one pixel, a function of sensor, focal length, and altitude. For ±2 cm vertical accuracy you want a GSD of 1.5 to 2.5 cm/px, which for a common 20 MP mechanical-shutter sensor means flying 60 to 90 m above ground. Halving altitude improves GSD but doubles flight lines; the planning question is always the accuracy the decision actually needs. Enforcement evidence for unauthorized water capture needs survey grade; a monthly erosion trend can fly higher and faster.

Overlap is cheap insurance

Photogrammetry reconstructs surfaces from features seen in many frames. Our minimums: 80 % forward overlap, 70 % side overlap, raised to 85/80 over water edges, dams, and vegetation where features are weak. Overlap costs battery, and it is the single cheapest way to avoid re-flying a site because a model would not resolve. On steep dam faces, add a second battery for oblique passes at 30 to 45 degrees; nadir-only imagery systematically underestimates vertical structure.

RTK gets you close, control points make it defensible

An RTK/PPK drone tags each frame with centimeter-grade coordinates and removes most geolocation error. It does not remove the need for independent verification. Our standard: at least 5 ground control points for absolute registration on first surveys of a site, then 3 or more checkpoints (points withheld from processing) on every survey, purely to measure error. The checkpoint RMSE is the number you quote when someone challenges the volume, and "the drone has RTK" is not an accuracy statement anyone should accept.

Conditions that quietly ruin models

  • Wind above 10 m/s degrades mechanical stability and effective overlap.
  • Low sun (under about 25 degrees) casts long shadows that read as false relief in the surface model.
  • Moving water cannot be reconstructed; plan water-body surveys around the shoreline and use the water surface elevation from a staff gauge, not the model.
  • Uniform surfaces (fresh asphalt, calm silt) starve the matcher of features; oblique light helps, crossed flight lines help more.

From frames to numbers

Upload the mission and Examin runs the photogrammetric chain (alignment, dense reconstruction, DEM and orthomosaic generation) and then the vision core takes over: volume computation against a reference surface, change detection against the previous survey, and anomaly flags (new intake structures, erosion features, encroachment) with coordinates. Each deliverable carries its checkpoint RMSE, so every number ships with its own error bar.

The mission-plan template

For a repeatable site: fixed takeoff point, fixed altitude and overlap, same time-of-day window, GCP monuments that survive between surveys, and a checkpoint set that never feeds processing. Change detection is only as honest as the consistency of the surveys being compared, and consistency is a planning decision, not a processing option.

Related modules

Deployment window open

Put your cameras to work.

A 30-minute walkthrough on your own footage. See what the vision core finds on day one.