Creating three-dimensional thermal maps

Abstract

A method for generating 3D maps of mines with thermal imaging texture from a robotic platform is presented. The objective is to use these models to generate risk maps that can be used to assess mine safety. Analysis will take place offline, but the data collection platform will be autonomous. Our registration method is based on aligning 3D descriptors that are extracted from range images and uses ICP for refinement. The descriptors encode the distribution of radial distances in the vicinity of keypoints, and enable alignment of non-sequential scans that are visibly different. Thermal texture is added to the registered point cloud using a once-off calibration between the 3D and thermal cameras. Two alternate visualisation schemes are discussed, and results are shown for a real mine stope.