Coordinate Systems & Georeferencing

A coordinate system is the agreement that turns numbers into positions: an origin (where zero is), a set of axes (which directions the numbers count along), and a unit. Give a point three numbers and a frame, and you have located it unambiguously. Change the frame and the numbers change while the point stays put. This is the first thing to internalise, because almost every alignment headache in reality capture is really a frame mismatch in disguise.

In practice you work in one of two broad kinds of frame. A local or project coordinate system is one you invent for the job: put the origin at a convenient corner of the site, run the axes along the main building grid, measure in metres, and record everything relative to that. It is simple, intuitive, and perfectly sufficient when all you need is the internal geometry of one building — the relationships between its own walls, floors and openings. A phone scan of a room, a standalone photogrammetry model, a quick as-built: these often live happily in a purely local frame, sometimes even an arbitrary one the software picks automatically.

A real-world (geographic or geospatial) coordinate system locates points on the earth itself — by latitude, longitude and height, or by a grid of eastings and northings — so that your data sits in the same frame as maps, site surveys, GIS layers and other projects. You need the real-world frame the moment your capture must relate to anything outside its own four walls: the legal site boundary, the neighbouring plots, a topographic survey, municipal data, a masterplan, or another discipline's model that is itself georeferenced. The cost of a real-world frame is that it drags in the complications of the next two sections — the earth's shape and how we flatten it — which is exactly why you do not reach for it unless the job needs it. The discipline is to choose deliberately: use the simplest frame that serves the purpose, record clearly which frame your data is in, and never assume two datasets share a frame just because their numbers look similar. A local model dropped unthinkingly next to a georeferenced one will sit in the wrong hemisphere, or a few hundred kilometres out, and the cause is always the same — the frames were never reconciled.

The moment you want real-world coordinates, you collide with two facts: the earth is not flat, and it is not even a tidy sphere. It is an irregular, slightly flattened, bumpy shape, and to give positions on it we need two agreements — a datum and a projection — each of which is a modelling choice with real consequences.

A geodetic datum is the reference model of the earth that your coordinates are measured against: a mathematically defined ellipsoid (a smoothed, slightly squashed sphere) positioned and oriented to fit the earth, either globally or for a particular region. Global datums such as the one underlying GPS are designed to fit the whole planet reasonably; regional and national datums are tuned to fit one area especially well. The crucial practical point is that the same physical spot has different coordinates on different datums — potentially differing by many metres — so latitude and longitude are meaningless unless you also know the datum. Mixing datums, or assuming two datasets share one, is a classic way to end up metres out while every individual number looks perfectly reasonable. In India, national survey and mapping sit within the Survey of India framework, and getting the datum right for survey-grade work is precisely the kind of thing you defer to licensed surveyors rather than guess at.

A map projection is the second agreement: the recipe for flattening the curved reference surface onto a flat plane so we can draw maps and work in simple planar x, y coordinates. Here there is an unavoidable truth worth carrying for life: no flat map can preserve shape, area, distance and direction all at once — flattening a curved surface always distorts something. Every projection therefore makes a deliberate trade, keeping some properties faithful at the expense of others, and is accurate only over the region it was designed for. Widely used systems such as the Universal Transverse Mercator grid divide the world into zones precisely to keep distortion small within each zone. For a designer the takeaways are not the mathematics but the judgement: know that a datum and a projection together define a real-world coordinate system; know that distortion and datum differences are real and can reach metres; record exactly which datum and projection your data uses; and treat the setting-up of survey-grade georeferenced coordinates as professional survey work, governed by the national framework and verified against control.

If real-world coordinates need a frame tied to the earth, something has to supply position in that frame, and for most modern work that something is GNSS — Global Navigation Satellite Systems, the umbrella term for the satellite constellations that let a receiver compute where it is. GPS is the best-known, alongside other global systems and India's own regional constellation; a GNSS receiver can use several together. The principle is elegant: each satellite broadcasts its position and a precise time; the receiver measures how long the signals took to arrive from several satellites at once, converts those times into distances, and solves for the one position consistent with all of them. Position from timing and geometry — measurement again, indirect as ever.

What matters for capture is that GNSS accuracy spans an enormous range depending on technique. A bare consumer receiver — the one in a phone — typically places you within a few metres, fine for navigation and rough context, useless for survey. Survey-grade methods that use correction data from reference stations and carefully held observations can reach centimetre-level, and are the normal way a surveyor fixes control points in real-world coordinates. The gap between those regimes is vast, and confusing them is a serious error: a phone's position is not a surveyed coordinate, and treating it as one will place your carefully captured building several metres from where it really sits. GNSS also has real limitations — it needs a clear view of the sky, so it struggles or fails indoors, under dense canopy, and among tall buildings where signals are blocked or bounce, which is exactly why indoor and enclosed captures rely on local control rather than GNSS alone.

The practical role of GNSS in reality capture is usually not to position every point of the scan directly, but to establish a few accurate real-world positions — control points, scanner locations, or targets — into which the dense local capture is then fixed. The scanner measures the building precisely in its own local frame; GNSS (or a total station tied to GNSS control) tells you where that local frame sits on the earth; and georeferencing, the subject of the final section, marries the two. As always, the binding version of this — survey-grade GNSS control, real-world coordinates that others will rely on legally or structurally — is the work of licensed surveyors and verified equipment, operating within the Survey of India and the governing standards, not something to improvise from a handheld reading.

Georeferencing is the act that ties it all together: assigning real-world coordinates to your captured data so that it sits correctly in the earth's frame and agrees with every other dataset referenced to that frame. Concretely, you take a capture that was made in a local frame and compute the transformation — a shift, a rotation, a scaling — that moves it into a real-world coordinate system, so that the corner of the building that the scanner called (12.4, 8.0, 2.6) now also carries its true position on the earth. Georeferencing is what lets your point cloud line up with the site survey, the property boundary, the topographic map, the GIS layers and the other consultants' georeferenced models, all in one coherent space.

The usual method is clean in principle: measure a set of control points whose real-world coordinates are known to good accuracy (from GNSS or a total station tied to control), identify those same points in your capture, and solve for the transformation that best brings the two into agreement. With enough well-distributed control points, the software aligns the whole dataset to the world, and the leftover mismatches at the control points tell you how well it worked — a built-in check on the quality of the georeferencing. This is why control, the subject of the next lesson, is the hinge between local accuracy and real-world placement: without it, a capture can be internally perfect and globally lost.

Why does this matter enough to be its own lesson? Because so much of capture's value comes from integration. A georeferenced scan can be compared to the design model to verify construction; combined with neighbouring surveys to understand a site; fed into a GIS or a city model; revisited years later and overlaid on a new capture to detect change; and trusted to coordinate with work done by people you will never meet. An ungeoreferenced scan is an island — useful for the building's own internal geometry, but unable to speak to anything outside itself. That said, know the boundary firmly. Accurate, survey-grade georeferencing — the kind that underpins legal boundaries, statutory maps, setting-out, and deformation monitoring — is licensed survey work, carried out to recognised standards, within the national framework (in India, the Survey of India), against proper control, and certified by professionals. This course teaches you what georeferencing is, why it matters, and how to specify and sanity-check it; it defers the binding establishment of real-world coordinates to the surveyors and the verified procedures whose job that is. Know which frame your data is in, insist that captures which must integrate are properly georeferenced, and hand the survey-grade version to the professionals.