From Raw Drone Imagery to Deliverable: The Photogrammetry Pipeline Explained

Clients see two moments: the drone taking off, and the finished deliverable landing in their inbox. What happens between those two events is where survey quality is actually determined — and where most providers cut corners.

This is the unedited version of our photogrammetry pipeline.

Stage 1: Mission Planning

Photogrammetry accuracy begins before the drone leaves the ground. The capture pattern, altitude, overlap percentages, and GCP (Ground Control Point) layout are calculated based on the required GSD (Ground Sampling Distance) and project deliverable specifications.

For standard topographic mapping at 2cm GSD, we fly at 80m AGL with 80% frontal overlap and 70% lateral overlap. For sub-centimeter deliverables, we reduce altitude to 40m and increase overlap to 90%×85%. The math matters: insufficient overlap means reconstruction gaps. Excessive overlap means processing time that drives cost without proportional quality gain.

GCP placement is the single most important factor in absolute accuracy. We deploy a minimum of eight RTK-surveyed GCPs per project, distributed at corners and interior positions, verified to ±1cm. The control network is the foundation everything else is built on.

Stage 2: Raw Data Acquisition

The Altis M2 Mapper captures 61-megapixel images at defined intervals throughout the flight plan. A 100-hectare survey generates approximately 2,400 individual frames, each geotagged with RTK-corrected position data accurate to 2cm.

Image quality is non-negotiable. Blur, poor exposure, and haze degrade tie-point matching in processing. We fly in optimal lighting windows — 2 hours after sunrise and 2 hours before sunset — and scrub flights where atmospheric conditions fall below our image quality threshold. Raw data that enters the pipeline at 95% quality cannot produce a 98% quality deliverable.

Stage 3: Structure-from-Motion Processing

SfM (Structure from Motion) is the computational engine that reconstructs 3D geometry from 2D images. The algorithm identifies thousands of matching keypoints across overlapping images and uses their parallax — the apparent movement of objects against each other from different viewpoints — to calculate depth.

This produces a sparse point cloud, essentially a skeleton of the scene’s geometry. We run this on our GPU processing cluster, which compresses what would be 12 hours of desktop computation into 90 minutes for a standard 100-hectare project.

The sparse cloud is then densified using Multi-View Stereo (MVS) algorithms, which evaluate every pixel across every image to add color and geometry detail. A 100-hectare project typically generates 800 million to 1.2 billion dense points.

Stage 4: Quality Inspection and GCP Registration

Before any outputs are generated, every project goes through a mandatory quality inspection. We check:

• Point cloud coverage: No holes or data-sparse zones in the project area
• GCP residuals: Individual GCP errors must be below 2cm. Projects exceeding this threshold are flagged for re-flight
• Check point accuracy: Independent GCPs not used in registration confirm absolute positional accuracy
• Texture quality: Dense cloud texture resolution validates input image quality

If anything fails inspection, the project stops here. We re-fly before we deliver.

Stage 5: Deliverable Generation

Approved point clouds are the source for every downstream deliverable:

Orthomosaic: A geometrically corrected, top-down aerial image where every pixel has a known geographic coordinate. Resolution matches the calculated GSD. Output formats include GeoTIFF with full CRS metadata, ready for any GIS platform.

Digital Elevation Model (DEM): A raster surface representing terrain elevation. Used for slope analysis, drainage modeling, cut/fill calculations, and flood risk assessment.

3D Mesh: A textured polygonal model of the surface, exportable as OBJ, FBX, or DAE for use in CAD and visualization platforms.

Point Cloud: The raw LAS/LAZ file for clients who require direct point cloud integration with their software environment.

The 48-Hour Commitment

From landing to delivery is 48 hours, guaranteed. Our processing infrastructure runs continuously, and our QA team reviews every deliverable before it leaves the building. The client receives a structured delivery package: files organized by type, a quality report with residual statistics, and a coverage map confirming 100% project area capture.

The data doesn’t just arrive. It arrives ready to use.