TBBR: Annotated RGB and Thermal Drone Images of Building Rooftops

On March 19, 2019, a drone flight captured 926 aligned RGB and thermal images of building rooftops in Karlsruhe, Germany. The dataset includes 6,927 polygon annotations identifying thermal bridges, with each structure annotated multiple times from different angles due to high image overlap. It also contains a height map, providing a multimodal view for analyzing building heat loss.

Use Cases

• [Training object detection models] based on [6,927 polygon annotations of thermal bridges]
• [Multimodal fusion analysis] based on [aligned RGB, thermal, and height map data]
• [Urban energy audit simulation] based on [thermal imagery of building rooftops captured under controlled morning conditions]
• [Dataset augmentation research] based on [high overlap rate where each structure appears ~20 times from different angles]

Strengths

• Contains 6,927 detailed polygon annotations of thermal bridges, providing precise ground truth for model training.
• Images are multimodal, consisting of aligned RGB, thermal, and height data captured under consistent environmental conditions (temperature: 3.78-4.97°C, humidity: 80-98%).
• High image overlap provides multiple views (~20 on average) of each annotated thermal bridge, useful for robustness testing.

Limitations

• Sources conflict on the exact image count (926 vs. 917) and annotation count (6,927 vs. 6,895), and on the final cropped image dimensions (2680x3370 vs. 2400x3400).
• Data is from a single one-hour flight on one morning, limiting temporal and weather condition diversity.
• Metadata such as column names, exact file formats, and last update date are not provided.

Provenance

Source

Zoe Mayer, Karlsruhe Institute of Technology

Collection Method

Recorded with RGB and FLIR-XT2 thermal cameras on a DJI M600 drone, annotated using VGG Image Annotator (v2.0.10).

Time Range

March 19, 2019, 7:00-8:00 AM

Geography

City centre of Karlsruhe, Germany, east of the market square, covering six building blocks.