Spatial AI · Digital Twin · CityGML Scene Graph
I'm a researcher at AI Digital Twin Lab, Dong-A University.
My research focuses on graph-based spatial intelligence systems that connect
3D city/building models, visual observations, semantic-spatial relations, and digital twin data.
I am currently working on transforming CityGML building models and image-based scene observations
into structured graph representations for spatial reasoning, querying, localization, and mapping.
-
AI Digital Twin Lab, Dong-A University
- Researcher, M.S. Student in Computer Engineering
- Research areas: Spatial AI, Digital Twin, CityGML Scene Graph, Smart City, Graph-based Spatial Reasoning
-
ORCID
- Spatial AI
- Digital Twin
- CityGML / 3D City Models
- Scene Graph / Knowledge Graph
- BIM · GIS · Smart City
- Vision-Language Model based Spatial Understanding
- Neo4j-based Graph Querying and Spatial Reasoning
- Sensor Data Integration for Building Operation
flowchart LR
A[Input Image] --> B[View-Aware Graph]
C[CityGML Model] --> D[CityGML Scene Graph]
D --> E[Neo4j World Graph]
B --> F[Graph Matching]
E --> F
F --> G[Localization & Mapping]
E --> H[Spatial Querying]
E --> I[Digital Twin Services]
I am currently working on a spatial reasoning pipeline that connects:
-
Input Image → View-Aware Graph
Extract visible objects and viewpoint-aware spatial relations from a single image using a VLM. -
CityGML → CityGML Scene Graph
Convert semantic 3D building and city models into graph structures. -
Scene Graph → Neo4j World Graph
Store semantic-spatial relations in a graph database for querying and reasoning. -
View Graph ↔ World Graph Matching
Compare observed view graphs with existing world graphs for localization and mapping.
| Project | Description | Keywords |
|---|---|---|
| 3DCitySG | A Python-based research framework for constructing semantic-spatial scene graphs from CityGML building models. | CityGML, Scene Graph, Neo4j |
| View-Aware-Graph | A framework for extracting structured viewpoint-aware graph JSON from a single input image using a VLM. | VLM, View Graph, Spatial AI |
3DCitySG focuses on converting CityGML building models into explicit semantic-spatial scene graphs.
Main features:
- CityGML 2.0 building model parsing
- Building, BuildingPart, Room, BoundarySurface, Opening, and BuildingFurniture extraction
- Semantic hierarchy construction
- Geometry-aware spatial relation modeling
- AABB-based spatial relation inference
- Neo4j persistence for graph-based querying and analysis
View-Aware-Graph focuses on extracting structured graph representations from a single urban scene image.
Main features:
- Single image input
- VLM-based scene object extraction
- Viewpoint-aware spatial relation extraction
- JSON schema-based output structure
- Object and relation confidence representation
- Downstream connection to graph matching, localization, and mapping
Spatial AI
Digital Twin
CityGML
Scene Graph
Knowledge Graph
Neo4j
Cypher Query
BIM / GIS
Smart City
Sensor Data
SensorThings API
FROST Server
Vision-Language Model
Frontend Development
Backend Development
Android Development
Jetpack Compose
MVVM
REST API
Docker