AI4TWINNING
Subproject: Thermal 3D mapping and CNN analysis
Project Investigator
Prof. Dr. Ludwig Hoegner,
Chair of Photogrammetry and Remote Sensing
Project contributors
Manoj Kumar Biswanath, M.Sc.,
Chair of Photogrammetry and Remote Sensing
Project summary
- Development of an extended 3D thermal description of buildings by combining indoor and outdoor thermal mapping, in order to establish a 3D spatial reference for temperature values of building facades which has advantage over traditional ways of manually inspecting thermal images with regard to energy consumption and their efficient usage.
- Analysis of thermal patterns such as inhomogeneous distribution of wall materials, leakages, heating pipes and anomalies, by using techniques such as segmentation and semantic interpretation.
- Implementation on TUM city campus buildings and performance evaluation of the methods using an annotated model.
Project (preliminary) results
3D thermal point cloud is projected to facades of building model. An algorithm searches for nearest neighbor points in point cloud to map the thermal intensities to generate the thermal textures. The method showed advantages such as detection and reduction of occlusions as well as window detection and removal from the textures for better temperature estimation of the façade walls.