Camera settings for capture

Coverage and overlap decide whether a scene can be reconstructed at all; camera settings decide whether the individual photos are sharp enough for COLMAP to actually use that coverage. A perfectly planned capture path shot at the wrong shutter speed still fails, just later in the pipeline, when feature matching quietly drops the blurred frames or, worse, keeps them and produces soft, ghosted geometry. The settings below aren’t about getting a nicer-looking photo; they’re about giving structure-from-motion the sharp, consistent input it needs.

Shutter speed: the one setting that isn’t negotiable

Every other setting in this article involves a trade-off. Shutter speed mostly doesn’t: a blurred frame contributes little to a reconstruction regardless of how well everything else was set. Drone-mapping guidance converges on a version of the same rule stated a few ways: keep motion blur during the exposure under about one pixel of ground-sample distance, ideally half a pixel of GSD, which translates in practice to shutter speeds of 1/500s or faster for typical mapping flights, never below 1/200s when flying slow or high, and never below 1/500s when flying fast or low. One frequently cited practical starting point for drone mapping is 1/800s. The same logic carries over to handheld or walking ground capture, just without the GSD maths: at typical walking pace and normal focal lengths, a comparable practical floor is somewhere around 1/250-1/500s, not the 1/60s or 1/30s you might get away with for a single static tripod shot. If you’re capturing while moving at all, walking a room, flying an orbit, treat shutter speed as the setting you fix first and build everything else around.

ISO and aperture: the trade-off underneath

With shutter speed pinned to whatever motion in the shot demands, ISO and aperture are what’s left to balance exposure. Aperture guidance for photogrammetry-style capture generally lands around f/8-f/11: narrow enough to get meaningful depth of field so more of the scene stays in focus, but not so narrow that diffraction starts softening detail, which becomes a real cost past roughly f/16. ISO should stay as low as the remaining exposure budget allows, commonly cited ranges are up to around 800 on full-frame sensors and lower, around 500, on crop sensors before noise starts working against feature matching. When all three settings are in tension, the practical ordering that shows up repeatedly in capture guidance is shutter speed first, then ISO, then aperture: a sharp but slightly noisy frame still registers reliably, while a smooth, low-noise frame that’s motion-blurred usually doesn’t.

Motion blur as the real enemy

It’s worth being explicit about why motion blur outranks noise as a problem, because they look similar in a quick preview but behave very differently in a matcher. Noise is high-frequency but roughly random from frame to frame; the underlying edges and corners a feature matcher locks onto are still there, just with some grain sitting on top, and matching degrades gracefully. Motion blur smears exactly those edges and corners into a gradient, removing the fine structure the matcher needs rather than just adding noise on top of it. A moderately noisy but sharp set of photos will usually reconstruct; a clean, low-noise set with even mild motion blur across a chunk of the capture routinely produces gaps, mismatched registrations, or blurred geometry in exactly the areas that were smeared.

RAW vs JPEG

RAW capture gives real recovery latitude, particularly for exposure and white balance, which matters most in the high-dynamic-range conditions interiors with windows or facades in direct sun tend to produce. The cost is storage, transfer time, and an extra conversion step, since COLMAP and training tools consume standard image formats, JPEG, PNG, or TIFF, not camera RAW directly, so a RAW-based workflow means batch-processing every frame to a consistent output before it ever reaches COLMAP. If lighting across the whole capture is already even and there’s no expectation of recovering blown highlights or crushed shadows after the fact, JPEG shot straight out of camera is a reasonable, faster choice. For particularly demanding jobs, some photogrammetry software vendors specifically recommend exporting to TIFF rather than JPEG, on the basis that it avoids JPEG’s lossy compression artefacts feeding noise into the reconstruction. The decision is really about how much post-capture correction you expect to need, not about raw quality in the abstract.

In practice

Pick a shutter floor for the session based on how fast you’re actually moving through the scene, drone orbit or room walkthrough, and don’t drop below it even when a scene gets dim; drop ISO or open aperture instead, or add light. Shoot in aperture-priority or full manual with ISO capped rather than full auto, so exposure doesn’t drift between frames the way exposure lockFixing exposure, and ideally white balance, to one value for an entire capture pass rather than letting the camera re-meter every frame. Prevents auto-exposure from making the same surface look different across photos, which otherwise confuses feature matching. on a locked walkthrough pass is meant to prevent. And if shooting RAW, batch-convert to a single consistent JPEG or TIFF profile before handing anything to COLMAP, mixed formats or per-photo edits introduce exactly the kind of frame-to-frame inconsistency that undoes careful capture planning further up the pipeline.