The Reality Capture Landscape

Almost every way of capturing reality into 3D data belongs to one of three families, distinguished by *how* they measure.

Laser scanning (LiDAR) is an active method: the instrument emits its own light — laser pulses — and measures how those pulses return, timing them or analysing the returning wave, to compute the distance to whatever they hit. Sweep that measurement across millions of directions and you directly produce a dense, metrically accurate point cloud. Because the instrument supplies and measures its own signal, laser scanning is typically the most directly accurate family, works in the dark, and handles large distances; it is also, at the high end, the most expensive. This is the subject of Module 3.

Photogrammetry is a passive method: it uses ordinary photographs and infers 3D geometry from them. Take many overlapping photos of a subject from different positions, and software finds common features across the images, works out where each camera was, and triangulates the 3D position of the points — the same parallax principle your two eyes use for depth. Photogrammetry needs no special emitter, just cameras, so it is cheap and accessible and produces rich colour and texture; but it depends on good lighting, texture and photo technique, and struggles with blank or shiny surfaces. This is Module 2.

Neural capture is the newest family: Neural Radiance Fields (NeRF) and Gaussian splatting use machine learning to build a 3D scene from photographs. Rather than solving explicitly for point positions, they learn a representation of the scene that can render strikingly photorealistic new viewpoints. They excel at visual realism and at tricky materials that defeat classical photogrammetry, but their geometry can be metrically unreliable, and the field is evolving fast. This is Module 4.

Two things are worth fixing in mind now. First, the families are not rivals so much as complements — real projects routinely combine them, for example a laser scan for metric accuracy plus photogrammetry or neural capture for rich visuals. Second, all three converge on the same kind of output: measurable 3D data — a point cloud, a mesh, a model — that downstream work, including scan-to-BIM, can use. Different roads, same destination; the art is knowing which road suits the journey.

Rather than memorise a table of methods, it is far more useful to understand the *axes* along which they all vary, because every real choice is a trade-off among them. Five matter most.

Accuracy — how closely the captured data matches the true geometry. This ranges from survey-grade (millimetre-class, on proper control) down to the rough-but-useful. Crucially, accuracy is not a property of a 'method' in the abstract but of a method *plus its setup* — a careful photogrammetry job can beat a careless scan.

Cost — of the equipment, the software, the skilled operator and the processing time. This spans from lakhs of rupees for a professional terrestrial scanner to nearly free on a phone you already own.

Speed — how quickly you capture, and how long the data then takes to process. Some methods capture fast but process slowly (photogrammetry can need heavy computation), others the reverse. Consider both halves.

Range and scale — how far the method reaches and how big a subject it suits. A phone is happy with a room; a terrestrial scanner with a building; a drone with a site or landscape; each becomes impractical far outside its natural scale.

Portability and access — how easily you can get the kit to the subject and operate it there, including regulatory access: drone flight, for instance, is regulated.

The central truth of the whole field is that *you cannot maximise all of these at once.* Push for the highest accuracy and you usually pay in cost, speed or portability. Choose the cheapest, fastest, most portable option and you typically give up accuracy, range or robustness on difficult surfaces. There is no single 'best' method — only the method best matched to a particular job's requirements and constraints. This is why the professional question is never 'what is the best scanner?' but 'what does *this* job need, and what is the cheapest, fastest way to meet that need reliably?' Throughout the course you will see these trade-offs recur; learning to reason along these five axes is what turns a list of gadgets into real judgement. And one axis sits above the others as a discipline: whatever you choose, you must be able to state the accuracy you are getting — and defer any binding figure to verified specifications and, where it matters, a licensed surveyor.

The method families are realised in a spectrum of hardware, and it helps to place the main types along it from the most capable-and-costly to the most accessible.

At the top sit survey instruments and terrestrial laser scanners (TLS) — tripod-mounted scanners, often paired with total stations and control, producing the highest accuracy over buildings and large interiors. They are the workhorses of serious building survey, costing lakhs, needing skilled operators, and capturing from fixed setups that are later registered together. This is the professional core for accurate as-builts.

Next come mobile and handheld scanners using SLAM — Simultaneous Localisation And Mapping — which let you *walk through* a space while the device continuously builds the map and tracks its own position. SLAM trades some accuracy for enormous gains in speed and convenience: you can capture a whole floor in a walk-through rather than dozens of tripod setups. Ideal where speed and coverage matter more than the last millimetre.

Drones (UAVs) carry cameras or LiDAR into the air, making them the natural choice for sites, roofs, facades and landscapes — anything large or hard to reach from the ground. They unlock aerial photogrammetry and aerial LiDAR, but their flight is regulated (in India, under the Drone Rules and the DGCA framework), which is part of choosing them.

At the accessible end is the phone in your pocket. Modern phones with LiDAR sensors, plus free or low-cost photogrammetry and scanning apps, can capture rooms and objects genuinely usefully — enough for many interior jobs, quick as-builts, visualisation and, above all, learning. They do not give survey-grade accuracy or handle large, complex buildings well, but they put real capture in everyone's hands.

The pattern across the spectrum is consistent with the trade-offs of the previous section: as you move from tripod scanner to phone, cost and accuracy generally fall while speed, portability and accessibility rise. No single device is 'right'; the right device is the one whose place on the spectrum matches the job. For an Indian context where professional kit is concentrated in larger firms and metros while capable phones are everywhere, this spectrum is not academic — it is exactly the practical menu from which real projects choose. Module 7 treats hardware and method choice in depth.

So far everything produces *captured data* — a point cloud or mesh that faithfully records measured reality. But a point cloud, for all its density, is in an important sense *dumb*: it is millions of undifferentiated points. It knows nothing of 'wall' or 'window' or 'column'; it cannot tell you a door's fire rating or a beam's size; you cannot schedule quantities from it directly. It is measured reality, not understanding.

Scan-to-BIM is the step that adds the understanding. It is the process of turning the captured point cloud into a structured BIM model — replacing the cloud of points with intelligent, named, parametric objects (a wall that knows it is a wall, a window that carries data) positioned to match the captured reality. This is where measured geometry becomes a usable, data-rich model the whole team can design, coordinate and schedule against. It is the second half of this course (Module 6), and it is worth being honest now about a fact we return to in the next lesson: scan-to-BIM is still largely a *manual, interpretive* task — a skilled person modelling objects to fit the cloud — with automation emerging but imperfect. Capture gives you measured reality cheaply; turning it into a model still takes human judgement and time.

How, then, should you begin to think about choosing across this whole landscape? Work backwards from the decision the data must support. Ask: *What will this capture be used for, and to what accuracy must that use be reliable?* A space-planning sketch of a simple interior, a conservation record of a heritage facade, a structural intervention, and a run of prefabricated joinery have wildly different needs. Then ask about *scale and access* (a room, a building, a site; reachable on foot or only from the air), about *budget and time*, about whether you need *metric accuracy, visual realism, or both*, and about whether the result must end as a *BIM model* — which adds the scan-to-BIM effort on top. The answers point you to a region of the landscape, not a single product. And one question frames all the others: does any part of this cross into binding territory — boundaries, setting-out, monitoring, survey-grade georeferencing — in which case it belongs to a licensed surveyor, working to verified specifications and the governing rules. Read the map first; the later modules teach you to walk each road.