Capturing Good Photos

If you remember one rule, make it overlap. Because the software reconstructs a surface point only from the parallax between photos that both contain it, every point you care about must appear in *several* overlapping images. The common guidance is that neighbouring photos should share 60 to 80 percent or more of their view - and on curved, complex or featureless surfaces you push that higher, not lower. Overlap is cheap insurance: photos cost nothing, holes and failed builds cost a return visit. When in doubt, take more, closer together.

Overlap has to be *two-dimensional* on a broad surface. It is not enough to shuffle sideways along a wall in one line; you also need overlap between rows as you move up and down, so think of your photos as tiling the surface like overlapping shingles, each frame sharing generous margins with the ones beside, above and below it. For an object or a room you achieve the same thing by orbiting: circle the subject in small, even steps so each photo overlaps the last, then repeat the circuit at a higher and a lower angle so tops, sides and undercuts are all seen from more than one direction.

Coverage is overlap's partner: the discipline of leaving no gap and no surface seen from only one angle. A surface captured from a single direction cannot be triangulated and will be weak or missing, so every important face needs views from at least two, ideally several, directions. Work systematically - a planned orbit, grid or series of rows - rather than wandering and shooting at random, which reliably leaves holes you discover only after processing. Pay special attention to transitions and recesses: inside corners, reveals, the undersides of sills and cornices, the backs of columns. These are both the most occluded and the most useful for an as-built, so slow down and add extra, angled shots there. And add detail passes: after the broad coverage, move in close on ornament, junctions and anything intricate, so those areas get the higher image density they need. Capture is not a few clever shots; it is a methodical sweep that guarantees every surface is seen, sharply, from several sides.

Photogrammetry matches features by how a surface *looks*, so anything that makes a surface look different from one photo to the next, or gives it nothing to look at, is the enemy. Start with light. You want even, diffuse, consistent illumination across the whole capture - an overcast day outdoors, or soft, uniform artificial light indoors - so that a given patch of wall looks the same from every angle. Harsh direct sun is a classic trap: it throws hard shadows and blown highlights that move as you move, confusing the matching, and the shadows themselves get baked into the model. Avoid capturing while the light is *changing* (passing clouds, the sun tracking across a facade, someone flicking lights on and off), because inconsistent lighting between photos is as damaging as poor lighting. Crucially, photogrammetry reads the light that is *on* the surface - it does not strip shadows out - so shadows and glare in your photos become defects in your model.

Then there are the surfaces that lie or say nothing. Three families cause most failures. First, featureless surfaces - a plain painted wall, a smooth ceiling, clear sky, fresh snow - give the software nothing distinctive to match, so they reconstruct as noise or holes. Second, reflective and shiny surfaces - polished metal, gloss paint, mirrors - show a different image from every viewpoint, so the 'features' move and the geometry is wrong. Third, transparent surfaces - glass, water, clear plastic - let the camera see through or into them, so the software tries to reconstruct reflections or whatever lies behind. Mirrors are the worst of both worlds, fabricating a whole false room. The practical responses: for blank surfaces, add temporary texture where acceptable (projected patterns or removable markers) or accept that you will measure those planes another way; for glass and shiny things, expect gaps and plan to model them as simple planes rather than trusting the cloud; and everywhere, capture on a cloudy day or with soft light to tame reflections. Finally, avoid motion - people walking through, leaves and traffic, a swaying tripod - and anything that changes between shots, because the scene must be effectively static across the whole set. Knowing these failure modes in advance turns them from baffling bad results into predictable, plannable constraints.

Because matching depends on fine detail, sharpness is sacred - and a blurred photo is worse than a missing one, because it injects wrong information. The goal of your settings is simple: every photo crisp, correctly exposed, and consistent across the set. If your camera allows manual control, the photogrammetry habits are: a reasonably small aperture (a higher f-number) to get enough depth of field that the whole subject is in focus, not just one plane; a low ISO to keep noise down, since noise masquerades as fake texture and muddies matching; and a shutter speed fast enough to freeze any shake or motion - which usually means bracing the camera, using a tripod, or shooting in good light so you can keep the shutter quick.

Consistency across the set matters as much as any single setting. Lock focus and exposure where you can, so the software is comparing like with like rather than photos that jump in brightness and focus. Shoot at the highest quality and resolution your storage allows - and if your camera offers RAW, it preserves the most information for careful exposure correction later - because more, finer detail means more features to match and better accuracy. Keep the lens clean, and avoid extreme zoom settings and heavy distortion where you can; most software calibrates the lens from the photos, but clean, consistent optics make its job easier and the result more reliable.

Two practical cautions. First, avoid anything that alters the image after capture in ways the software cannot model - aggressive in-camera filters, panorama stitching, heavy HDR, or editing individual photos differently - because photogrammetry wants honest, consistent frames, not artistically processed ones. Second, mind the trade-off between quantity and manageability: more overlap is good, but thousands of redundant near-identical frames slow processing enormously for little gain, so aim for *deliberate* coverage - enough overlap and angles to see everything well, without drowning the software. For a phone, the same principles apply through whatever pro or manual mode it offers: lock focus and exposure, shoot in good even light, hold it steady, and keep the subject filling the frame. Good settings will not save bad coverage or bad light - but bad settings will ruin good coverage, so treat sharpness and consistency as the floor everything else is built on.

From Lesson 2.1 you know the sting in the tail: photographs recover shape but not absolute size, so a capture intended for *measurement* must include known scale. In practice that means placing one or more scale references in the scene and keeping them in frame across enough photos - a calibrated scale bar, a printed target of precisely known length, or at minimum a ruler or an object whose true dimension you have measured. Use several references spread through the capture, not a single one in a corner, so the scale is anchored across the whole model and you can cross-check it. For work that must tie to real-world coordinates, or whose accuracy must be checkable and defensible, you move up to surveyed control points whose coordinates a surveyor establishes - and that, along with any georeferenced or binding deliverable, is the licensed surveyor's domain, not an app's automatic scale.

That leads to a repeatable capture routine worth internalising. *Before:* scout the subject, check the light (pick an overcast window or arrange even artificial light), clear or note movers and clutter, and lay out your scale references. *Settings:* set a small-ish aperture, low ISO, fast-enough shutter, highest quality, and lock focus and exposure. *Capture:* do a systematic broad pass - orbit or rows with heavy overlap in both directions, every surface from several angles - then a second and third pass at different heights, then close-up detail passes on ornament, junctions and recesses, keeping the scale references visible throughout. *Check on site:* review a few frames for sharpness and exposure, confirm you have covered the tricky corners, and measure one or two real dimensions with a tape so you can verify the model's scale later. *After:* back up the photos before you leave.

The overarching mindset is to capture for the software's sake, not the eye's. A beautiful photograph and a useful photogrammetry frame are different things: the software wants many honest, sharp, overlapping, evenly-lit images that between them see every surface from several sides, with known scale in view - redundancy and discipline, not artistry. Build that routine into a habit and the hard part of photogrammetry is done before you ever open the software; skip it, and no processing will save you. And always verify: measure against a tape, and hand anything accuracy-critical or binding to a licensed surveyor.