Automated field mapping

On this page 11

The base layer everything else sits on

One flight produces a high resolution 3D model of the field: elevation, drainage lines, headlands and problem ground.

CategoryDetect
Slugfield-mapping
CadenceOnce per field, then refreshed after drainage work or a boundary change
PayloadHigh resolution RGB, RTK, LiDAR on wooded or tall canopy blocks
Needs a visit firstNo

The problem

Most farms are still working from a field boundary drawn once and a memory of where the wet spot is. Without an elevation model there is no way to see why one corner floods, where the water actually runs, or how much of a field is not worth cropping at all.

How it works

StepWhat happens
SurveyThe drone flies an overlapping photogrammetry grid over the whole block
ReconstructImages are matched into a point cloud, then a surface model and an orthophoto
DeriveSlope, aspect, flow accumulation and depressions are calculated from the surface
PublishLayers are written to your field record so every later flight is measured against the same base

What you get

  • Orthophoto
  • Digital surface and terrain models
  • Drainage and flow layer
  • Corrected field boundary with area

What the dashboard measures

  • Mapped area
  • Elevation range across the block
  • Depression area and volume
  • Boundary area against the registered area

That last one pays for the flight on its own more often than people expect. A registered area and a measured area disagreeing by three percent is common, and it has been quietly driving every per-hectare figure on the farm.

Where it matters most

Winter wheat, sugar beet, grassland and silage, cooperatives, reforestation.

Build it

This is the capability to build first. The payload is one ordinary camera, the entire pipeline is open source, and every other capability is measured against the base layer it produces.

Flight parameters that actually matter

ParameterValueWhy
Forward overlap75 to 80%Below 70% the reconstruction starts failing over uniform canopy
Side overlap65 to 70%
Altitude80 to 120 m AGL2 to 3 cm/px with a 24 MP APS-C sensor and a 24 mm lens
SpeedSuch that motion blur stays under half a pixelAt 1/1000 s and 10 m/s that is fine; at 1/250 s it is not
PatternDouble grid on tall or wooded blocksSingle grid is enough for bare or short canopy
Ground controlRTK on the aircraft, plus 3 to 5 checkpointsCheckpoints are how you find out the RTK was lying

Fly with the shutter as fast as light allows, and prefer a global shutter or a mechanical shutter. A rolling shutter on a moving aircraft skews every frame, and photogrammetry software has to model that skew rather than measure ground.

Hardware

ItemWhyEU sourceUS sourceIndicative
Airframe, 5 to 7 kg classCarries a real camera for 30 minutesHolybro X500 V2 kit via Drone Parts CenterGetFPV, RobotShop€400 to €600
Flight controllerArduPilot or PX4Pixhawk 6Xsame€250 to €330
RTK GNSSCentimetre camera positions, which removes most ground controlArduSimple simpleRTK2BSparkFun ZED-F9P€200 to €300
Mapping cameraThe one part not to economise onSony ILX-LR1, or a used a6000 or a7R IIB&H€400 used to €3,000 new
Lens, 24 mm fixedFixed focus, no zoom, no stabiliserKamera ExpressB&H€200 to €500
Camera triggerShutter fired from the flight controller, geotagged from the logArduSimple accessoriesSparkFun€40
Checkpoint targetsPainted or printed ground targets, surveyed onceEmlid Reach RS3 to survey themsame€2,500 for the receiver, targets are paint

Budget alternative for the camera: a Raspberry Pi Global Shutter camera (BerryBase, The Pi Hut, about €55) plus a C-mount lens. It is 1.6 MP, so you fly lower and longer for the same ground resolution, and the reconstruction is noisier. It is genuinely usable for drainage mapping and genuinely not usable for anything that needs plant-level detail.

In-house software

StageWhat we runLicenceReplaces
Flight planningCustom planner emitting MAVLink missions, or QGroundControlBSD / GPLDJI Pilot, Pix4Dcapture
GeotaggingMatch shutter events in the flight log to framesown codevendor tools
ReconstructionOpenDroneMap via NodeODM, or COLMAP plus OpenMVSAGPL / BSDPix4Dmapper, Agisoft Metashape, DJI Terra
Raster workGDAL, rasterio, PROJMIT / BSD
Terrain derivativesWhiteboxTools, pysheds, or GRASS r.watershedMIT / GPLArcGIS Spatial Analyst
Boundary extractionVectorise the orthophoto edge, snap to the registered parcelown code
ServingCloud Optimized GeoTIFF plus TiTilerMITMapbox tiling

The whole chain runs unattended. One 25 hectare block at 2.5 cm/px is roughly 600 frames, about 40 minutes on a mid-range GPU box with ODM.

Licence note worth knowing before you build a product on it

OpenDroneMap and WebODM are AGPL-3.0. If a customer interacts with them over a network, the AGPL's network clause applies to your modified version. Two clean ways out: run ODM as a batch job whose outputs you serve, and keep it unmodified, or build on the permissive stack instead, COLMAP (BSD) plus OpenMVS (AGPL, so check that too) or OpenMVG (MPL-2.0). Get this decided before it becomes 40,000 lines of integration.

See Ingest and photogrammetry for the pipeline in detail.

Cost efficiency

  • Do not buy an RTK base. Use a network correction service. Several German states publish SAPOS corrections free of charge, and Centipede RTK and RTK2go cover much of Europe for nothing. A base station is for sites with no mobile data.
  • Do not pay per hectare for stitching. This is the single largest recurring cost in commercial drone mapping and it is a solved open source problem. One GPU box pays for itself in the first few thousand hectares.
  • Free elevation data first. Several German states publish a 1 m DEM as open data. If a customer only needs drainage lines on flat arable land, the state DEM may answer the question without a flight. Sell them the flight when they need centimetres, not when they need metres.
  • Spend on the camera and the shutter. Everything downstream inherits their quality, and nothing downstream can recover it.