Generierung von thermischen Punktwolken von Gebäuden aus Infrarotbildsequenzen und MLS-Punktwolken

Thermal infrared imagery (TIR imagery) is widely used to assess building energy use and leakage. To reduce the difficulties of image interpretation and occlusion in TIR images, the transition from image data to georeferenced 3D coordinates of buildings is required. In this paper, point clouds of mobile laser scanning (MLS) are automatically enriched with thermal intensities from a TIR image sequence. First, a method for extracting and mapping salient points from both the point cloud and TIR images based on line inter-sections is presented and compared with the extraction of 2D and 3D Harris Corner points. Then, a correspondence determination algorithm based on RANSAC is proposed and tested for the orientation estimation of the TIR images relative to the point cloud. The point cloud is then projected into every image to interpolate TIR intensities for every point. Finally, Non-Local-Means is used to determine a final intensity value for each 3D point from all TIR images where the 3D point is visible.