Drone-to-BIM Automation

Ohio (USA) -
Efficient BIM Model Creation for Large Areas

•Task: Automate the creation of a BIM model (LOD 300) for a 1000 m² area using drone video, reducing time and manual effort while improving scalability and management efficiency

• Large-scale construction projects, renovation planning, urban development, BIM model creation

Our Technology

Lepei technology enables to create high-quality BIM models from VIDEO with LOD 200 accuracy in just 1-2 hours, saving up to 80% of resources compared to the traditional Scan to BIM process.

Fast result:
Drone Video
BIM Model

*We are currently working on upgrading the model to LOD 400, which will help us create even more detailed and accurate models.

**The speed and accuracy of the model depends on the specific types of objects, see the diagram below

Taxonomy for the industrial use case: The complete point cloud is separated into object classes, that are divided into sub-classes over three levels of granularity.

Working Process:
1. Scan: Drone Video

High-resolution 4K 60 FPS video footage is captured to ensure maximum coverage of the construction site.

GPS and IMU data are extracted for georeferencing, enabling accurate positioning in the final BIM model.

2. Dense Point-cloud
Structure-from-Motion (SfM) and Multi-View Stereo (MVS) algorithms reconstruct a dense, high-fidelity point cloud from video frames.

Neural Radiance Fields (NeRF) are used to refine point cloud density and enhance depth accuracy.
3. Semantic Point-cloud (different color per object type)
Deep learning segmentation models (e.g., PointNet++) classify point cloud data into distinct object categories (walls, floors, vegetation, furniture, etc.).

Color-coded layers are assigned based on object type, improving visualization and downstream processing.
4. Object layers (exp: vegetation)
Edge detection and feature recognition algorithms separate distinct elements

Graph-based clustering techniques extract surfaces, enabling precise identification of elements required for BIM modeling.
5. Object clustering (exp: vegetation)
Machine learning algorithms (DBSCAN, RANSAC) refine object clustering, removing outliers and improving model accuracy.

Geometric simplification techniques optimize the dataset for efficient processing in BIM environments.
6. Object processing → Revit
Automated Revit API integration converts the cleaned and structured point cloud into a fully functional BIM model.

LOD 200+ object classification ensures the model is structured with predefined walls, doors, slabs, and mechanical elements in Revit.

Conclusion and Benefits for you Business

Conclusion and Benefits for you Business

Faster BIM Creation – Reduce modeling time from days to hours, accelerating project timelines.

Significant Cost Savings – Save up to 80% of manual labor and scanning costs compared to traditional Scan-to-BIM workflows.

Scalability – Automate BIM generation for large-scale projects without additional fieldwork.

Increased Accuracy – AI-driven processing improves model consistency and reduces human errors.

Future-Proof – Ongoing improvements to LOD 400 will enhance detail for high-precision architectural and engineering needs.


For more information on the technologies used, please follow the link

Want to see this in action?

Contact Us

Contact Us

Contact Us