Drone surveying, mission plan to deliverable

A photogrammetric mission is the same shape every time, whether you're flying ten acres or a thousand. The decisions you make in stage one cascade through the other three — bad overlap planning at the desk shows up as holes in the dense cloud two days later.

Photogrammetry recovers 3D geometry from where the same feature appears in multiple photos. The more photos a point appears in, the better the bundle adjustment locks it down. Overlap is how you make sure that happens.

• Front lap (along-track overlap). Percent of each photo that overlaps with the next photo on the same flight line. 75% is the common floor for survey-grade work; bump to 80–85% over dense vegetation or repetitive textures.
• Side lap (between flight lines). Percent of overlap between adjacent passes. 65% is the floor; 70–75% for tall vegetation or building corridors where occlusions are common.
• Why both matter. Front lap alone gets you 2D coverage; only the combination of front + side lap gives the redundancy a multi-image bundle adjustment needs to resolve elevation.

GSD is the on-the-ground length of one pixel. It's the single number that decides whether you can resolve a curb edge, a pothole, or a fenceline.

Two corollaries drop out of the formula. First: doubling altitude doubles GSD — you cover twice the ground per photo, but each pixel grows. Second: a longer focal lens at the same altitude gives you finer GSD at the cost of footprint. Surveyors usually want the lowest altitude the project tolerates and a fixed focal length matched to the sensor.

Rule of thumb: expected horizontal map accuracy is about 1–2 × GSD on an RTK drone with ground control, and 3–5 × GSD without it. Plan altitude so the achieved GSD is half of the smallest feature you need to identify.

A GCP is a surveyed point with a high-contrast target painted or laid on the ground. Photogrammetry uses GCPs two ways: as control (the bundle adjustment is forced to honor their coordinates) and as check points (independent points whose residuals tell you the achieved accuracy).

• Layout. One GCP at every corner of the project plus one in the middle. For projects longer than ~500 m in any dimension, add a row through the middle every ~300 m. The math behind this: GCPs constrain the bundle, but residuals grow with distance from the nearest constraint.
• Survey them. RTK rover with at least 30 seconds of observation; longer occupations and a second epoch some hours later for the highest order work. Record the controller's reported HRMS / VRMS for each GCP — the bundle adjustment can use those as input weights.
• Targets. 0.5 m × 0.5 m chequerboard or chevron, high contrast, matte (not glossy). Centre over the surveyed point and photograph it so the target appears in at least 4–5 images.
• Checks. Hold back ~20% of your points as check points. After processing, the RMS of their residuals is what you report as the project's achieved horizontal and vertical accuracy.

Photogrammetry software (Pix4D, Metashape, RealityCapture, WebODM) runs the same general pipeline regardless of vendor. Understanding it helps when one stage fails — the right error to fix is usually the stage before the obvious one.

• Alignment. Feature matching across photos and a sparse bundle adjustment. If alignment fails on a subset of photos, overlap is too low or texture is too uniform — not a processing problem.
• Dense matching. Per-pixel depth from the aligned photos. Slow; this is the stage where overnight processing happens.
• Mesh / DSM. Triangulate the dense cloud into a surface model. DSM = digital surface model, includes trees and buildings; DTM = digital terrain model, ground only (filter step required).
• Orthomosaic. Each pixel of each input photo, reprojected through the DSM onto a flat coordinate grid. Saves to GeoTIFF for hand-off into CAD or GIS.

• Wind kills sidelap. Plan with the published wind tolerance of your aircraft, not the still-air spec. A 15 kt quartering wind on a Mavic 3 Enterprise eats 8–10% of side lap by the time you cross-track the project. Bump planned side lap to compensate.
• Sun angle and shadows. Best flight window is 2 hours either side of solar noon. Avoid early/late shadows that the bundle adjustment can't match between photos — same place, different shadows = no feature match.
• Water and snow. Reflective, texture-less, often featureless from above. Either avoid them in the flight footprint or accept that you'll have holes in the deliverable over them.
• Battery temperature. Cold batteries fail in flight without warning. Pre-warm to 20°C before takeoff in sub-freezing conditions; expect 25–30% reduced flight time anyway.