From Flight to Map: How a Drone Survey Actually Works

Every survey begins with a brief: what do you need the data for? The answer determines every technical parameter that follows — resolution, accuracy class, deliverable format, and whether the job calls for a drone at all. A site boundary for a developer needs different resolution to a topographic model for a dam engineer. A volumetric calculation for a quarry has different accuracy requirements to a preliminary site layout for a residential estate.

Once the brief is clear, a desk-based site assessment reviews the area using satellite imagery, terrain data, and airspace maps. In Kenya, this means checking KCAA (Kenya Civil Aviation Authority) Class G airspace restrictions, proximity to controlled airspace around Wilson and JKIA, restricted zones, and the altitudes permitted without special authorisation. A NOTAM (Notice to Airmen) is filed for the survey period. Any site within 5 km of a controlled aerodrome requires prior written KCAA approval — this takes time and must be factored into the project schedule, not discovered on the morning of the flight.

A physical site visit before the flight day confirms what satellite imagery cannot: access routes for the ground team, the location of power lines and tall trees that constrain take-off and landing zones, the terrain relief that determines safe altitude above ground level, and any site-specific hazards — active construction, livestock, security restrictions, community sensitivities — that affect survey operations.

Mission planning is where the brief becomes a flight programme. The four interdependent parameters that drive all mission design are Ground Sampling Distance (GSD), flight altitude, image overlap, and flight line spacing. GSD is the size of one pixel in the final orthophoto when measured on the ground — a 3 cm GSD means each pixel represents a 3 cm square on the ground. GSD is determined by the flight altitude: higher altitude = larger GSD = less resolution. For a DJI Phantom 4 RTK at 100 m AGL, the GSD is approximately 2.7 cm/pixel. At 200 m, it doubles to approximately 5.4 cm/pixel.

Image overlap — the percentage of each photo's area that appears in adjacent photos — is the second critical parameter. Survey-grade work requires a minimum of 75% forward overlap (between consecutive photos along a flight line) and 65% side overlap (between adjacent flight lines). These minimums ensure that every point on the ground appears in at least three photos from different angles — the redundancy that photogrammetric processing requires to compute reliable 3D positions. Operators who reduce overlap to cover more ground faster are trading data quality for time. The penalty shows up in the accuracy report, not the preview.

The mission is programmed into flight planning software — DJI Pilot 2, Mission Planner, or Pix4Dcapture — which calculates the exact flight line layout, turn radii, camera trigger intervals, and estimated flight time. For sites with significant terrain relief, terrain-following mode maintains constant AGL altitude across the survey area, keeping GSD consistent even as the ground rises and falls.

Ground Control Points (GCPs) are the stage that separates a survey-grade drone deliverable from a visually appealing but geometrically unreliable map. A GCP is a physical target — typically a 50×50 cm cross or chequerboard pattern, either pre-painted on the ground or a purpose-made target board — placed at a known, precisely measured location. The photogrammetric software uses GCPs to anchor the 3D reconstruction to real-world coordinates, eliminating the accumulated position and altitude errors that even the best drone GNSS introduces over a flight.

GCP positions are measured using a survey-grade GNSS receiver (RTK or post-processed static) tied to a base station or CORS (Continuously Operating Reference Station) network. The measurement must be precise: errors in GCP coordinate measurement propagate directly into the accuracy of the entire final map. A GCP measured with a hand-held GPS receiver accurate to ±5 m provides essentially no geometric control — it is no better than the drone's onboard GNSS. Survey-grade GCP measurement achieves ±2–5 cm in horizontal position and ±3–8 cm in vertical elevation.

GCP distribution across the survey area matters as much as the accuracy of individual measurements. The standard pattern requires GCPs at all four corners of the survey area, at the midpoints of the longer edges, and one or more in the interior — a minimum of five for a simple rectangular area, with additional GCPs wherever terrain or land cover changes significantly. Additionally, Check Points (CPs) — independent survey points that are not used as inputs to the photogrammetric processing — are measured and withheld, to be used only for independent accuracy verification after processing is complete.

With mission loaded, GCPs in position, and airspace clearance confirmed, the drone launches. For a 10 km² survey at 100 m AGL with 75%/65% overlap, a single battery will cover approximately 2–3 km², meaning three to five battery cycles for the full mission. The drone follows its programmed flight lines autonomously, triggering the camera at calculated intervals. The pilot monitors the mission in real time — watching for battery drainage, unexpected airspace intrusions, signal interference, and any anomalies in the flight path logged on screen.

Weather is the variable most underestimated by clients scheduling drone surveys. Wind above 8–10 m/s (Beaufort 5) affects image quality and positional accuracy, particularly for light multi-rotor UAVs. High cloud cover reduces contrast and degrades feature matching in photogrammetric processing. Direct overhead sun creates harsh shadows that obscure ground features. The ideal window in Kenya is typically early morning — 7:00 to 10:00 AM — before the convective heating of the day generates thermal turbulence and afternoon cloud build-up over the highlands. Professional survey teams have weather holds built into their project schedules; clients who push for a fixed delivery date regardless of weather create conditions for poor data.

After each flight, the pilot performs an immediate data card check — confirming photo count matches the expected total, reviewing sharpness and exposure on a sample of images, and confirming GCPs are visible and correctly positioned in images taken near each target. If any flight line produced blurred or underexposed images, it is re-flown on the same day before the team de-mobilises. This immediate QC on site is far less costly than discovering missing data during office processing.

Photogrammetric processing is where hundreds of individual photographs become a spatially referenced 3D model. The process runs in specialist software — Pix4Dmapper, Agisoft Metashape, or DJI Terra — and proceeds through three sequential computational stages that build on each other.

Aerotriangulation (AT) is the first and most fundamental step. The software analyses keypoint features in every image — corners, edges, texture patterns — and identifies thousands of matching points that appear across multiple overlapping photos. By knowing the camera's focal length, sensor size, and the approximate position of each photo from the drone's onboard GNSS, the software computes the precise position and orientation of every camera at the moment each photo was taken. The GCP coordinates are then introduced, anchoring the entire bundle of images to the real-world coordinate system and eliminating the accumulated drift that GPS-only positioning produces. The AT stage produces a sparse point cloud — the mathematical skeleton of the survey.

Dense image matching (MVS — Multi-View Stereo) takes the calibrated camera positions from AT and reconstructs a dense 3D point cloud by matching every pixel across multiple overlapping images. For a 500-photo survey, this can produce 50–200 million points representing every visible surface in the survey area. Processing time depends heavily on hardware — a workstation with a high-performance GPU reduces processing of a 500-photo survey from 24 hours to 2–4 hours. The dense point cloud is the raw material from which all subsequent products are derived.

The Digital Surface Model (DSM) is generated by gridding the dense point cloud at the specified resolution — interpolating the highest point in each grid cell to produce a continuous raster surface representing the top of everything: ground, vegetation, buildings, vehicles. In open terrain without trees or buildings, the DSM closely approximates the bare-earth Digital Terrain Model (DTM). Where vegetation or structures are present, the DSM and DTM diverge — the DSM shows the top of the canopy or roof, while the DTM requires additional filtering and classification to extract the underlying ground surface.

The orthophoto — or orthomosaic — is the map product most clients recognise. It is a planimetrically corrected aerial image of the survey area: unlike a conventional photograph taken at an oblique angle, an orthophoto has been geometrically corrected so that every point is in its true horizontal position, free of the perspective distortion that makes buildings lean and ground closer to the camera appear larger than ground further away. Measurements taken from an orthophoto — distances, areas, angles — are geometrically accurate to the precision of the underlying point cloud and GCP network.

The orthophoto is generated by projecting each input photograph onto the DSM surface — the software drapes each image over the 3D model in its correct geometric position, then blends the thousands of individual photos into a seamless mosaic. Colour balancing adjusts exposure and white balance across the mosaic to remove the visible seams that would otherwise appear where photos taken at slightly different times, sun angles, or exposure settings meet. A well-balanced orthophoto is indistinguishable from a single photograph of the area at the survey resolution; a poorly balanced one shows visible grid lines and colour shifts across photo boundaries — a telltale sign of either poor overlap or inadequate processing.

Before any data is delivered to the client, the accuracy of the outputs is verified against the independently measured Check Points withheld in Stage 3. The software computes the Root Mean Square Error (RMSE) between the Check Point coordinates as measured on the ground and the same points as they appear in the orthophoto and DSM. This RMSE — in centimetres, separately in horizontal (X, Y) and vertical (Z) — is the honest measure of the delivered data's real-world accuracy. A well-executed survey with properly measured GCPs and adequate overlap typically achieves horizontal RMSE of 1–2× GSD and vertical RMSE of 1–3× GSD.

The accuracy report is the professional certification that accompanies every Geopin survey deliverable. It documents the GCP and CP measurement method, the number and distribution of control points, the processing software and settings, and the achieved RMSE against Check Points. This report is what engineers, planners, and developers need to confirm the data meets the accuracy specification in their brief — and it is what separates a professional survey deliverable from a consumer drone flyover.

Final deliverables are compiled in the client's required coordinate system — typically Arc 1960 / UTM Zone 37S for Kenya — and exported in the required formats. The complete data package is transferred to the client with a handover note explaining what each file is, how to open it, and the accuracy limitations that apply to the dataset.