HIDA: Tackling Heat Leaks with Data Science

This post is written by HIDA and originally appeared on HIDA’s website.

At the HIDA Hackathon on Computer Vision, data scientists met in Berlin to develop innovative solutions for reducing energy losses in buildings.

From 27 to 29 February 2024, the Helmholtz Information & Data Science Academy and HEIBRiDS organized the HIDA Hackathon on Computer Vision at the Einstein Center for Digital Future in Berlin. 14 doctoral candidates from the Helmholtz Information & Data Science Schools (HIDSS) took on a demanding challenge:

Using drone images taken over the rooftops of Karlsruhe, the data scientists were tasked with identifying and segmenting thermal leaks in houses. This is because thermal bridges – those poorly insulated areas of a house that conduct heat more quickly from the inside to the outside – often cause energy losses that account for up to a third of the transmission heat loss of an entire building. They also often lead to moisture and mold growth, damaging the fabric of the building.

The participants in the HIDA Hackathon were tasked with finding clever solutions to automatically detect thermal bridges in thermographic data. All approaches were allowed – from traditional image filtering techniques to semantic segmentation using deep learning. The annotated data for the hackathon was provided by host Markus Götz (KIT). It contained more than 70 GB of image data with five channels (RGB color, thermal image, height of buildings above ground) covering the rooftops of Karlsruhe city center. “I hope that we will see many different approaches to solutions, perhaps even something without AI,” explained Götz.

The hackathon gave participants the opportunity to demonstrate their data science skills and develop innovative solutions to a real-world energy efficiency challenge. The event was accompanied by mentors from various Helmholtz Centers and Helmholtz Imaging and offered the PhD candidates the opportunity to deepen their knowledge in the field of computer vision, exchange methods and gain valuable experience in the application of data analysis methods.

This hackathon was organized by KIT, DASHH, HDSLEE, HEIBRIDS, HIDSS, MARDATA, MUDS & Helmholtz Imaging.