Buildings invert the difficulty of the other 3D annotation verticals. The geometry is regular, the shapes are known, and what costs is not describing an object but deciding what it is among categories the trade defined long before the scanner arrived.This article sets out what this context demands. It extends the article on 3D annotation in forestry and agriculture.
What distinguishes this context
Five properties characterise this 3D annotation vertical and separate it from the preceding ones.The geometry is regular, planes, edges and right angles dominating the scene.The objects are standardised, a column or a duct belonging to known families.Density is high, a static survey producing a very dense cloud on the surfaces nearest the scanner.The deliverable is structured, a digital model expecting typed and related objects rather than a classification.And the conventions pre-exist, the building nomenclature being established independently of any annotation project.One important practical consequence follows. The fourth property distinguishes this vertical from all the others, the deliverable being not an annotated corpus but a structure a domain application can consume.The tasks requested
Five 3D annotation formulations occur in this vertical.Their cost differs strongly.Segmentation by element, which assigns each point to a wall, a floor or a ceiling.Detection of technical elements, ducts, cables, luminaires and equipment.Primitive extraction, a wall becoming a plane rather than a set of points.Structured reconstruction, which produces typed objects with their relations.And comparison with the design, which confronts the survey with the model that was expected.One observation follows. The third formulation is the tipping point, it turns an annotation into a geometric object and it constitutes most of the added value for use in a digital model.What regularity brings
Four exploitations follow from the geometry of buildings in 3D annotation.Plane fitting, a flat surface being determined on a set of points with high precision.The orthogonality constraint, a building’s walls mostly meeting at right angles.The verticality constraint, a wall and a column being vertical save documented exception.And repetition, a standard floor reproducing an organisation already annotated on the level immediately below.One practical consequence follows for a 3D annotation project. The fourth exploitation is the most worthwhile and the least used, a building with identical floors permitting an annotation to be propagated rather than redone, which divides the load on repetitive blocks.What site disorder imposes
This reality opposes the theoretical regularity.It dominates surveys during works.Four difficulties follow in 3D annotation.Clutter, materials and equipment masking the surfaces to survey.Temporary elements, props and scaffolding that do not belong to the structure.Unfinished surfaces, an unclad wall not presenting the geometry the finished building will have.And furniture in an occupied building, which masks floors and walls unpredictably.One important observation follows. The second difficulty calls for an explicit convention, a prop having to receive a dedicated class rather than being arbitrarily attached or omitted.The gap between survey and design
This frequent purpose changes the nature of the 3D annotation work.It deserves separate treatment.Comparing a built structure with its theoretical model produces conformity information.Three uses follow in 3D annotation.Execution checking, which verifies that an element is in place and to its dimensions.As-built recording, which updates the model to reflect what was built.And defect detection, settlement or deformation observed as a divergence from the original design.One practical consequence follows for a 3D annotation project. Those three uses presuppose a precise registration between cloud and model, a prior operation whose error carries fully into the divergences measured and merges with them.What the digital model expects
Four requirements distinguish a structured deliverable from a 3D annotation corpus.Typing, each object belonging to a category in the domain reference.Relations, a wall carrying an opening that carries a joinery item.Domain attributes, material, thickness and function.And parametric geometry, a wall being described by an axis and a thickness rather than by an envelope of points.One important observation follows for a 3D annotation project. The second requirement is the one annotation alone does not produce, a point-wise classification carrying no relation, which imposes a structuring stage distinct from the annotation work.The technical elements
This category concentrates the difficulty.It is often underestimated in 3D annotation.Services and equipment present four particularities in 3D annotation.Their fineness, a cable or a small duct being badly sampled by a survey designed for structural elements.Their entanglement, several services crossing within one plenum volume.Their diversity, a technical nomenclature counting far more categories than the shell.And their concealment, part of the services running behind closed surfaces or within slabs.One practical consequence follows. The last particularity bounds what a survey can establish, an embedded service not existing in the data, which must be declared rather than compensated by a reconstitution.The levels of detail expected
A graduation structures this vertical and it is standardised by the trade.Four tiers are distinguished by what they require in 3D annotation.The envelope alone, external walls, floors and roof, sufficient for a volumetric or energy use.The complete shell, partitions, columns, beams and openings included.The fit-out, joinery, finishes and fixed equipment.And the technical services, whose element count often exceeds that of all the preceding tiers taken together.One important practical consequence follows for a 3D annotation project. This level must be fixed by tier rather than by a general formula, a request for a complete survey with no level specified leading systematically to a scope disagreement at delivery.The adapted quality control
Five verifications exploit the properties of this 3D annotation vertical.The flatness of surfaces classified as walls, floors or ceilings.Orthogonality between elements, a large divergence signalling an error or a real defect.The verticality of load-bearing elements.The closure of volumes, a room having to be bounded by adjoining surfaces.And consistency between floors, a column having to be found at the same position from one level to another.One important observation follows for a 3D annotation project. The fourth verification is specific to this field and very effective, an unclosed volume signalling a missing or misclassified surface nothing else would reveal.Heritage and older buildings
A particular configuration deserves setting out since it contradicts the assumed regularity.An old building presents four divergences from contemporary construction.Walls are neither flat nor vertical, which invalidates the geometric constraints used elsewhere.Thicknesses vary along one wall, which makes representation by axis and thickness approximate.Angles are not right, which removes the orthogonality constraint.And the elements do not always fall into the categories a contemporary reference provides.One important observation follows for a 3D annotation project. Those four divergences mean a reference designed for new construction produces on an old building a model that smooths reality, which suits a volumetric use and misrepresents a heritage survey.What this work costs
Four factors determine the 3D annotation load in this vertical.The number of elements to type, the real basis of sizing.The level of detail expected, from the shell alone to the complete services.The clutter in the survey, which multiplies the time spent distinguishing the structure from the rest.And the structuring requirement, relations and domain attributes adding to the plain classification of points.One practical consequence follows. The second factor produces the largest variation, moving from the shell to the technical services multiplying the element count by a proportion clients rarely anticipate.The typing reference
One decision precedes any production and omitting it produces an unusable deliverable.Four elements compose that typing reference.The list of categories, drawn from a trade standard rather than built for the project.The attributes expected per category, with an indication of their origin, survey or documentation.The relations permitted between categories, a joinery item sitting in an opening that sits in a wall.And the target exchange format, whose exact version conditions what the client’s software will actually read.One practical consequence follows for a 3D annotation project. The last element is the one discovered latest, a deliverable conforming to the reference and exported in an incompatible version forcing a conversion part of the information does not survive.What repetition permits
One property of buildings deserves separate treatment since it constitutes the principal saving lever.A block frequently presents identical levels or identical dwellings.Four exploitations follow in 3D annotation.Propagating an annotation from one level to the next, with adjustment rather than rework.Cross-checking between levels, a divergence signalling either an error or a real singularity of the building.Building a model dwelling type, reused across all its occurrences.And concentrating human effort on the singular levels, ground floor, basements and top floors, where the repetition breaks.One important observation follows. The second exploitation is as useful as the first, a column absent at a single level being either an annotation omission or a structural particularity, two hypotheses the comparison flags and nothing else reveals.What pre-annotation contributes here
Three observations situate the contribution of 3D annotation assistance in this context.Plane detection works very well, the regularity of buildings being favourable.Automatic typing stays limited, a vertical plane potentially being a wall, a partition or a panel according to a context geometry does not carry.And structuring stays human, relations between objects belonging to an understanding of the building.One observation follows for a 3D annotation project. Those three findings lead to an effective division, assistance extracts the geometry and the human gives it a domain meaning, a division matching exactly this vertical’s value structure.What this vertical does not permit
Four limits bound what a survey can establish.What is embedded or cast in place does not exist in the data.The composition of a wall is not inferred from its surface.An element masked by furniture or storage stays inaccessible.And the state of a piece of equipment, working or out of service, is not read from its geometry at all.That second limit deserves emphasis. It means a material or thickness attribute comes from documentation or a probe rather than from the cloud, a distinction that must appear in the deliverable rather than blend into the other attributes.The first batch of a building project
Four zones compose a 3D annotation pilot batch that genuinely tests the conventions.A standard clear zone, which supplies the reference throughput.A cluttered zone, storage or furniture, which measures the effect of occlusion on the load.A plant room, which tests the services nomenclature and reveals the real element count.And a zone presenting an irregularity, a non-right angle or an out-of-plumb wall, which verifies that the geometric constraints are not being applied blindly to a building that does not respect them.One observation follows. The load gap between the first and the third zone is this batch’s most useful figure, it directly measures the effect of the level of detail and it supplies the basis for a scope discussion the floor area does not permit in 3D annotation.The uses that govern
Five purposes occur in this 3D annotation vertical, and they do not call for the same level of detail.Energy study, which needs surfaces and volumes and is satisfied by the envelope.Renovation, which needs the shell and the services likely to be affected.Operation and maintenance, which needs equipment typed and located.Execution checking, which needs divergences from the design rather than a complete model.And heritage survey, which needs the real geometry rather than an idealised one.One observation follows for a 3D annotation provider. The last purpose opposes all the others, it requires keeping the irregularities the first four require smoothing, which forbids handling a heritage survey with the conventions of ordinary construction.The conventions to write first
Five rules cover most of the 3D annotation disagreements in this vertical.The level of detail by tier, with the list of categories actually handled.The treatment of temporary elements, props and scaffolding receiving a dedicated class.The wall representation rule, axis and thickness or real envelope according to the purpose.What to do with occluded zones, declaration rather than reconstitution.And the declared origin of each attribute, survey or documentation.One practical consequence follows for a 3D annotation project. The third rule is the one that separates an ordinary use from a heritage one, and it must be decided before the first zone rather than discovered at delivery.Approaching a building project
Five questions scope a 3D annotation project in buildings.What level of detail is expected. That answer determines the element count and therefore the cost.Is the deliverable a classification or a structured model. That answer adds or removes an entire stage.Which typing reference applies. Its absence produces a nomenclature incompatible with the client’s tools.Is the survey in an occupied site or on a construction site. That answer predicts the clutter.And is a comparison with the design expected. That answer requires a prior registration.Those five answers determine the load and the feasibility. Asking them before starting avoids a deliverable the client’s domain software cannot read.The question that frames the project
One question determines the cost and the nature of the deliverable.Will the result be consulted by a human or consumed by software.A consultation-oriented answer, a virtual visit or a visual check, is satisfied by a classified cloud and places the project in the most economical regime.A software-oriented answer, a digital model or a calculation, requires typing, relations and a conforming format, which adds a structuring stage at least as expensive as the annotation.That question is asked in one sentence, it belongs to the use and not to the technique, and it avoids this vertical’s commonest misunderstanding, a client expecting a model and receiving a classification.Three decisions before the first zone
Three decisions commit the usability of a building 3D annotation deliverable.Fixing the level of detail by tier, with the named list of categories handled.Obtaining the client’s typing reference and the version of the target exchange format.And settling the wall representation, axis and thickness or real envelope, according to whether the purpose is ordinary or heritage.Those three decisions cost one meeting and a few file exchanges, they precede the first survey, and their absence produces careful work the client’s software cannot exploit.What this vertical brings to the others
Three practices born here hold beyond buildings.Propagating an annotation between identical configurations, applicable anywhere a structure repeats.Control by consistency between comparable units, transferable as soon as several instances of the same arrangement exist.And adopting a pre-existing domain reference rather than building a nomenclature, necessary whenever the deliverable feeds an established process.Those three practices presuppose a structural regularity few verticals offer, and their transfer depends less on the domain than on the presence of an exploitable repetition in the data itself. Where that repetition exists and goes unused, a project is paying twice for the same work.What this chapter teaches
One cross-cutting observation deserves closing this examination.This vertical moves the difficulty from geometry towards semantics.Three findings compose it.The regularity of buildings makes geometric extraction reliable and largely automatable, which removes the difficulty dominating the other verticals.Typing and relations belong to a domain understanding the data does not carry, which moves the value towards interpretation.And the reference pre-exists, which forbids inventing a nomenclature and requires adopting the trade’s.That finding matches the one the preceding chapter established about forestry: where the domain already holds its conventions, the work consists in conforming to them rather than in establishing them.Where the annotation work actually sits
One observation about scope belongs here, since this vertical is frequently misdescribed to providers.What is called 3D annotation in a building project is often three distinct pieces of work.Extracting the geometry, which is largely automatic and cheap per element.Typing the elements, which is human, fast per element and where the volume lies.And structuring the relations, which is human, slow per element and where the disagreements between client and provider concentrate.One practical consequence follows for a 3D annotation provider. Quoting these three as one line hides which of them the client actually needs, and separating them frequently reveals that a project asking for a full model needs typing everywhere and structuring only on the parts a calculation will consume.Common mistakes
These failures recur often enough that naming them is usually enough to avoid them.- Delivering a point-wise classification where a structured model was expected.
- Inventing a nomenclature instead of adopting the domain reference.
- Omitting a class for temporary site elements.
- Neglecting the prior registration in a comparison with the design.
- Presenting a material attribute as coming from the survey.
- Underestimating the move from the shell to the technical services.
- Expecting automatic typing to distinguish a wall from a partition.
- Omitting the volume closure check.
- Reconstituting an embedded service rather than declaring its absence.
- Ignoring the repetition between identical floors.
