3DONT

3D Ontologies for point cloud representation, management and querying

Re-imagining 3D Point Clouds

3DOnt introduces a groundbreaking approach to managing 3D point cloud data by integrating semantic technologies and knowledge graphs. Rather than treating 3D data as unstructured geometry, 3DOnt represents point clouds as 3D Graphs, enriching every point with meaningful, interpretable information. This enables professionals in fields like cultural heritage, smart cities, and environmental monitoring to explore, analyze, and understand 3D datasets through intuitive interfaces—no programming required.

Representing 3D data:

Ontology, Semantics and Mereology

Semantics at the Core

At the heart of 3DOnt is the “3D Graph”—a semantic, spatially structured representation that transforms raw point clouds into meaningful knowledge graphs. Each point is enriched with ontological concepts, allowing the system to understand and organize complex entities like “roof,” “column,” or “degradation risk.” This semantic layer doesn’t just describe what’s visible—it embeds contextual relationships, such as part-whole hierarchies or material dependencies, enabling automated reasoning and interpretation.

Smart, Open-Vocabulary Classification

3DOnt uses 2D imagery and projection techniques to automatically classify 3D point clouds with semantic labels from customizable ontologies. Its open-vocabulary system supports multi-label segmentation (points can belong to multiple objects) and adapts to different domains—no predefined class set or manual annotation required. This capability accelerates the structuring of 3D data and makes semantic enrichment scalable, explainable, and deeply integrable with downstream analysis.

Structuring Objects Through Mereological Relationships

3DOnt organizes 3D data into meaningful objects by leveraging mereological (part-whole) relationships encoded in domain ontologies. Instead of treating the point cloud as an undifferentiated mass, the framework identifies and links components—such as walls, beams, or columns—to their higher-level structures, like rooms or entire buildings. This semantic decomposition enables precise queries, targeted analysis, and reasoning over structural hierarchies. Users can explore not only where things are, but how they are composed and interconnected, unlocking deeper insight into the geometry and function of complex systems.

From Data to Knowledge:

Data Collection, Digital Twinning and Reasoning

Multimodal Data Integration through Referential Entities

The 3D Graph format allows for using entities - as nodes of the graph - as referential joints that anchor and unify all collected data across modalities. Each entity serves as a semantic hub, connecting geometric features, sensor observations, historical records, and inferred attributes within a coherent, queryable structure. This enables a multimodal and temporally-aware representation of reality, where heterogeneous data streams are continuously aligned to their referents. By organizing incoming data around ontological entities, 3DOnt ensures consistency, traceability, and semantic interoperability—laying the foundation for advanced reasoning, monitoring, and insight generation.

Heating Propension Index - calculated in 3DOnt by reasoning over:

Materials' Specific Heat Capacity
RGB

Flooding Risk Index - calculated in 3DOnt by reasoning over:

Materials' Permeability to Water
Geometries (Concavities: Steepness, Depth)
Proximity to Rivers

Explainable Reasoning for Insight and Automation

3DOnt integrates a powerful reasoning engine that goes beyond data visualization, enabling automated inference of high-level properties from geometric, semantic, and physical data. By leveraging the underlying ontological structure, the system can detect patterns, classify elements, infer missing relationships, and compute properties such as environmental risk. Reasoning in 3DOnt is transparent and explainable—every inference can be traced back to its sources and rules—giving users full control over how knowledge is derived and applied. This capability transforms raw 3D data into actionable insight, supporting decision-making, diagnostics, and intelligent monitoring.

Dew Formation Risk Index - calculated in 3DOnt by reasoning over:

Sensor-Measured Temperature
Sensor-Measured Humidity
Verticality
Proximity to Floor
Temporal Permanence of Critical Factors

Digital Twinning with Semantic-Geometric Intelligence

3DOnt redefines digital twinning by transforming 3D point clouds into rich, semantically structured digital replicas that evolve over time. Unlike conventional models, 3DOnt’s twins are built on a semantic-geometric foundation: every point is part of a knowledge graph that links spatial coordinates with domain-specific meaning, physical properties, and part-whole relationships. Real-time sensor data, 2D imagery, and historical records can all be integrated into this structure, enabling dynamic updates and multi-temporal analysis. Thanks to its open and explainable architecture, 3DOnt allows users to compute and monitor high-level properties—such as the dew formation index on surfaces—directly within the digital twin. These semantic twins are not only visually accurate, but also interpretable, queryable, and adaptable to changing conditions.

User-Friendly and Expressive Access to Data:

SPARQL-Based Natural Language Querying

Natural Language Querying

With 3DOnt, users can interact with complex 3D and semantic data through simple, natural language queries. Whether asking for "all surfaces with high dew formation risk" or "columns belonging to degraded buildings," the system parses the question, interprets its meaning using the underlying ontology, and returns precise results—visually and semantically. This intuitive interface opens advanced querying to non-technical users, bridging the gap between expert knowledge and raw data.

SPARQL for Expert Control

For advanced users and developers, 3DOnt offers full support for SPARQL—the standard query language for RDF graphs. This allows fine-grained access to the underlying semantic model, enabling complex data retrieval, filtering, and transformation. Users can write precise queries to extract, combine, and analyze spatial, temporal, and semantic information across the 3D graph, supporting advanced workflows, research, and integration with external tools.

3DGraphExplorer:

Interactive 3D View

The 3DOnt platform features a powerful web-based 3D viewer that seamlessly integrates geometry with semantic information. Users can explore annotated point clouds, inspect object hierarchies, view real-time sensor overlays, and highlight elements based on query results—all within an intuitive interface. Every point, object, or property in the viewer is linked to its semantic context, making the 3D model not just a visual reference, but a fully navigable knowledge space.

Embodied Graph: See the Structure in the 3D World

3DOnt can draw the knowledge graph inside the 3D scene itself, turning an abstract RDF structure into something you can orbit around. Each entity appears as a marker anchored at the centroid of the very points that constitute it; relations between entities stretch as color-coded arcs through space; the class hierarchy rises above the cloud as a tree. Every element is interactive: clicking an arc or a node reveals the relationships it embodies. A targeted SubGraph mode extracts only what matters—one relationship, one node, or their combination—while point-level relations are automatically attributed to the objects the points constitute, so the overlay stays readable at any scale. The graph is no longer beside the data: it inhabits it.

Taxonomy X Mereology: Dynamic Semantic Segmentation

Semantic segmentation in 3DOnt is not a fixed labeling but a live exploration across two ontological dimensions at once. The mereological axis controls which part level is displayed—from the smallest components up to entire entities—while the taxonomical axis controls how abstractly they are classified—from leaf classes up through their ancestors in the class hierarchy. The segmentation opens at maximum detail on both axes and is then re-shaped interactively with two sliders, recoloring the full point cloud in real time without re-querying the graph. The same scene can thus be read as "Pipelines, Solar Panels and Windows" one moment and "vegetation and buildings" the next, letting users find the exact semantic granularity their analysis needs.

Mereology at its Best

Part-whole hierarchies in 3DOnt are not just stored—they are a live control the user holds while exploring. Both Dynamic Instance Segmentation and Object-Level Property Plotting are governed by an interactive mereological level-of-detail slider that decides which object answers for each point.

Dynamic Instance Segmentation

In Dynamic Instance Segmentation, the mereological slider re-partitions the cloud on the fly: the same scene dissolves from whole buildings into their roofs and walls, down to the finest constituent components, one color per individual at the chosen level.

Object-Level Property Plotting

In Object Level Property Plotting, where values live on objects rather than points—a roof's surface, a building's height—the slider selects which of a point's nested carriers lends it its value, so the same predicate can be read per component, per part, or per entire entity. One slider, every scale of the part-whole hierarchy, recolored in real time.

Some Videos!

Related publications:

Codiglione, M., Beber, R., Remondino, F., 2024: Leveraging ontology for enhanced queries and analyses of urban point clouds. Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLVIII-2/W8-2024, 101–108
Codiglione, M., Mazzacca, G., Remondino, F., 2024: Dimensional discoveries: unveiling the potential of 3D heritage point clouds with a robust ontology framework. Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLVIII-2/W4-2024, 119–125
Codiglione, M., Remondino, F., 2025: NL-2-SPARQL: Ontology-Based Natural Language Querying over 3D Point CLoud Knowledge Graphs. Proc. Digital Heritage 2025, Eurographics Association
Codiglione, M., Remondino. F., 2026: 3D Semantic Digital Twins: Data Streams and Ontologies for Risk Prediction in Heritage Contexts. Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLVIII-2/W12-2026, 105-112

Page updated

Report abuse