The aim of the proposed project is to thermally record both the external and internal building surfaces and to spatially locate the measurements with the aid of a building model. Due to the fact that temperature is volume related, this strategy follows the idea “from surface to volume”. For this purpose, radiometric and geometric information will be recorded simultaneously at different times with a multi-sensor system consisting of a TIR camera and a laser scanner. Special challenges arise not only from the calibration, but also from the different conditions in the indoor and outdoor space regarding the resolution (multi-scale), the position and orientation determination, the co-registration and the corresponding analysis of spatial thermal structures. The relation between indoor and outdoor model shall be established by an automatic detection of windows and doors. Noticeable thermal structures which are difficult to describe by parametric models should be labeled and analyzed using a convolutional neural network (CNN). 

This is a subproject of the AI4TWINNING project funded by TUM Georg Nemetschek Institute Artificial Intelligence for the Built World

The project can be divided into two parts:

1. Create an extended 3D thermal description
2. Analysis of thermal textures