Capture Quality Starts Before the Button

Photogrammetry processing cannot recover detail that the camera never captured.

That makes image quality a mission-design problem, not a cleanup problem for later.

Before the first image exists, you have already made decisions about altitude, camera angle, route spacing, shutter behavior, lighting, overlap, scene motion, and the final product you expect to create.

Start with the required product

Suppose the stakeholder needs to identify roof seams and compare damaged areas.

That requirement affects capture choices.

A dataset designed only to create a visually pleasing overview may not preserve enough detail for the inspection claim.

Work backward:

stakeholder question
→ required visible feature / measurement
→ required image detail
→ camera geometry and altitude
→ route and overlap
→ capture settings / conditions

The camera button is near the end of the chain.

Ground sampling distance is sampling, not accuracy

A common mapping term is ground sampling distance (GSD): the approximate ground size represented by one image pixel under a given camera geometry and flight height.

Smaller ground area per pixel generally means finer spatial sampling.

But this statement is wrong:

My GSD is 2 cm, therefore my map is accurate to 2 cm.

GSD describes image sampling. Final positional or measurement accuracy also depends on:

• camera model;
• image sharpness;
• overlap;
• reconstruction geometry;
• geolocation/control quality;
• validation;
• scene characteristics.

A pixel size is not an error bound.

Blur destroys usable feature detail

Motion blur can come from:

• aircraft motion;
• vibration;
• slow shutter behavior;
• subject movement;
• poor stabilization.

Blur matters because the feature matcher needs identifiable edges, corners, and textures.

A blurry image may still look recognizable to a person and still be weak evidence for geometric reconstruction.

Exposure is an information problem

An image can fail because important detail is too dark or too bright.

If bright areas are clipped to white, the original texture may be gone.

If dark areas collapse toward black, useful features may disappear there too.

The goal is not one artistically perfect exposure. It is repeatable feature visibility across the dataset.

Large exposure changes between neighboring images can also make matching harder.

Lighting can change the scene

Shadows move.

Specular reflections change with viewing angle.

Wet surfaces can look very different from dry surfaces.

Clouds can change illumination during a long capture.

Those effects matter because photogrammetry works best when the same physical feature looks similar enough across views to match reliably.

The sun does not move the roof—but it can move the shadow that the software might mistakenly treat as a feature.

Altitude trades coverage against detail

Higher capture altitude generally increases the ground area visible in each frame for a given camera.

That can reduce the number of images needed for the same area.

It also usually makes each pixel represent a larger ground area, reducing fine detail.

Lower altitude can increase detail but may require:

• more images;
• more flight lines;
• more processing;
• more time;
• careful terrain/obstacle consideration.

The correct altitude is not “as low as possible” or “as high as possible.”

It comes from the product requirement, legal/operational limits, camera geometry, and site.

Oblique versus nadir imagery

Nadir imagery points generally downward and is common for orthophoto-style mapping.

Oblique imagery views the scene at an angle and can add useful geometry for vertical surfaces and 3D reconstruction.

Different workflows mix them differently.

OpenDroneMap documentation, for example, describes different overlap/camera-angle approaches for 2D/2.5D versus fuller 3D reconstruction.

Do not turn that into one universal capture recipe. Match camera angle to the product and scene.

Worked capture plan: simple field versus building

Site A — open athletic field

• mostly flat;
• strong painted-line features;
• few vertical structures;
• goal: orthophoto-style overview.

Site B — complex building

• roof edges;
• walls;
• vents;
• different elevations;
• goal: more complete 3D surface model.

A single identical flight/capture geometry is unlikely to be optimal for both.

The building may benefit from additional viewpoints or oblique geometry because vertical and complex surfaces need stronger observation from multiple directions.

Design for rejection margin

Assume some images will be bad.

A capture plan with barely enough overlap can collapse when several frames are blurred.

A stronger plan includes enough geometric redundancy that a few rejected images do not disconnect the observation network.

This is not permission to capture unlimited data. It is planned margin.

Build the capture-quality plan

For a supplied mapping scenario, define:

Then identify one choice that would change if the stakeholder asks for a more precise measurement product instead of a visual overview.

Misconception: processing settings can fix a weak image set

Processing can extract more or less from available evidence.

It cannot invent sharp texture, missing viewpoints, trustworthy scale, or stable scene geometry that was never captured.

The cheapest time to improve a dataset is usually before collection, while the capture plan can still change.

The Week 14 starting point

Capture quality is designed backward from the product:

claim → required detail → camera geometry → route/overlap → image quality → reconstruction.

Good photogrammetry starts before the shutter fires.