AI4BIM – Semantic Material Identification inside Elevator Shafts

Use Case:

Autonomous drone flights capture shaft imagery; an embedded YOLOv7 model classifies structural materials on the fly providing semantic maps for BIM updates renovation planning and safety checks.

Outcome:

• Technology elevated from TRL 5 → TRL 7 with a working MVP demonstrator
• Balanced dataset of 450 shaft images curated and fully annotated; dockerised training & inference pipeline for one-command deployment
• Robust real-time inference integrated into VERTLINER’s UAV software stack via Docker containers
• Risk of data scarcity mitigated through custom flights and synthetic augmentation, ensuring ≥80 % target accuracy despite limited brick examples

Ecosystem Support:

StairwAI vouchers funded two external AI experts who guided data strategy, annotation, YOLOv7 porting and hyper-parameter tuning; mentors from UNIBO provided technical oversightStairwAI vouchers funded two external AI experts who guided data strategy, annotation, YOLOv7 porting and hyper-parameter tuning; mentors from UNIBO provided technical oversight.

AI Relevance:

• Low-footprint, containerised AI deployable on edge hardware aboard drones.
• Demonstrates how SMEs can reach high-accuracy object detection with modest, domain-specific datasets.
• Combines AI with autonomous robotics to digitise hard-to-access built-environment assets.
• Opens a cloud-based revenue channel: subscription access to inspection datasets and BIM-ready semantic models.

Summary:

VERTLINER’s AI4BIM pilot transforms the traditionally hazardous and manual task of shaft inspection into an efficient, autonomous workflow. After collecting 450 high-resolution images via indoor UAV scans, the team—supported by external computer-vision experts—cleaned, balanced and densely annotated the data. A customised YOLOv7 network, fine-tuned through iterative data selection in CVAT / FiftyOne, now detects bricks, wood and steel beams with >90 % recall in under 50 ms per frame. The model ships as a Docker image, seamlessly plugging into the drone’s ROS-based flight software. Field pilots validated material-class bounding boxes with ≥70 % IoU against human labels, meeting StairwAI KPI thresholds. Thanks to this success, VERTLINER is commercialising AI4BIM within a broader service platform where contractors subscribe to inspection data and post-processing analytics—projecting entry into a construction-UAV market expected to reach €8.6 B by 2033.
AI4BIM exemplifies how targeted, explainable AI modules can enrich Building Information Models, enhance safety, and unlock new digital revenue streams for construction SMEs.