8.2 UAV Photogrammetry and Structure from Motion

UAV Photogrammetry & Structure from Motion (SfM)

The emergence of Unmanned Aerial Vehicles (UAVs or drones) combined with computer vision algorithms has revolutionized contemporary spatial data acquisition. In Nigeria, UAV photogrammetry has become an indispensable tool for cadastral surveys, corridor mapping, mining volumetric analysis, urban infrastructure planning, and disaster monitoring under regulatory frameworks set by SURCON and the Nigerian Civil Aviation Authority (NCAA).

Unlike traditional aerial photogrammetry which relies on heavy metric frame cameras with fixed flight geometries, UAV photogrammetry utilizes lightweight non-metric consumer or industrial digital cameras operating at low altitudes ($50\text{ m} - 150\text{ m}$). The geometric non-rigidity of consumer lenses is resolved using Structure from Motion (SfM) computer vision technology.

---

Structure from Motion (SfM) vs. Classical Photogrammetry

Structure from Motion is a digital imaging technology that simultaneously estimates 3D scene structure, camera positions, and orientation parameters from a series of overlapping 2D photographs. SfM fundamentally differs from classical photogrammetry in its mathematical approach to camera calibration and block adjustment.

The SfM Computational Workflow

1. Feature Detection and Matching: Computer vision algorithms such as Scale-Invariant Feature Transform (SIFT) detect robust image keypoints across multiple overlapping photographs regardless of scale, rotation, or lighting variations.
2. Sparse Point Cloud Generation: Feature correspondence across multiple photos is processed through relative bundle adjustment to reconstruct relative camera coordinates and a sparse 3D point cloud.
3. Self-Calibration Bundle Adjustment: The non-metric camera parameters ($f, x_0, y_0, K_1, K_2, P_1$) are calculated simultaneously during optimization.
4. Dense Multi-View Stereo (MVS): Dense matching algorithms compute depth maps for every pixel, generating a high-density 3D point cloud.

---

Ground Control Point (GCP) Layout & GNSS Integration

While SfM generates mathematically consistent relative 3D models, absolute georeferencing to national map coordinate systems (e.g., Minna Datum / UTM Zone 31N, 32N, 33N in Nigeria) requires Ground Control Points (GCPs).

GCP Distribution Strategy

• Perimeter Coverage: GCPs must be placed along the outer edges of the flight project boundary to prevent edge-model deformation (the "bowl effect").
• Internal Distribution: Control points must be distributed evenly across central areas and at terrain elevation extremes (ridge tops and valley floors).
• Check Points (CPs): A subset of independent ground survey points ($20% - 30%$ of total target points) must be withheld from the SfM bundle adjustment to serve as independent validation points for calculating Root Mean Square Error (RMSE).

RTK and PPK GNSS Drones

Modern survey-grade UAVs incorporate dual-frequency GNSS receivers onboard:

• Real-Time Kinematic (RTK): The drone receives real-time differential corrections from a local GNSS base station or Ntrip CORS network via telemetry link, recording phase-center camera exposure coordinates in real time.
• Post-Processed Kinematic (PPK): Satellite observation raw data is logged onboard the drone during flight and post-processed against GNSS base station data after landing. PPK is widely preferred in Nigerian field conditions because it avoids telemetry link dropouts caused by dense vegetation or topography.

Note: Even with RTK/PPK GNSS drones capable of $1-2\text{ cm}$ camera positioning accuracy, sound practice is to establish at least 4 to 5 independent ground verification checkpoints to confirm absolute vertical datum alignment.

---

Ground Sampling Distance (GSD) Mathematics

Ground Sampling Distance (GSD) is the ground distance represented by the distance between the centers of two consecutive image pixels. GSD dictates the spatial resolution and minimum feature size detectable on the orthomosaic.

$\text{GSD} = \frac{H \cdot s_w}{f \cdot i_w} = \frac{H}{f} \cdot \text{Pixel Size}$

Where:

• $H$ is flying height above ground level (AGL).
• $f$ is camera focal length.
• $s_w$ is physical sensor width (mm).
• $i_w$ is image width in pixels.
• $\text{Pixel Size} = s_w / i_w$ (micrometers or millimeters per pixel).

Example GSD Calculation:

For a UAV carrying a sensor with pixel size $3.45,\mu\text{m}$ ($0.00345\text{ mm}$) and focal length $f = 8.8\text{ mm}$, flying at $H = 100\text{ m}$ above terrain:

$\text{GSD} = \frac{100\text{ m}}{0.0088\text{ m}} \times 0.00000345\text{ m} = 11,363.64 \times 0.00000345\text{ m} = 0.0392\text{ m/pixel} = 3.92\text{ cm/pixel}$

---

Spatial Products: Orthomosaics, DSM, and DEM

Processing the dense 3D point cloud yields three primary spatial deliverables:

1. Digital Surface Model (DSM): A 2D raster grid representing the top reflective surface of the earth, including elevated natural and man-made features such as buildings, powerlines, trees, and canopy tops.
2. Digital Elevation Model (DEM) / Digital Terrain Model (DTM): A bare-earth surface model where non-ground objects (vegetation, vehicles, structures) have been digitally filtered out using classification algorithms (e.g., progressive cloth simulation or morphological filtering).
3. Orthomosaic: An orthorectified image product assembled from overlapping photographs where image displacements caused by camera tilt and terrain relief have been removed pixel-by-pixel using the underlying DEM. The orthomosaic possesses uniform scale across its entire extent, permitting direct planimetric distance and area measurements identical to a vector map.