A solution was developed that matches visible/IR imagery aboard a balloon in Saturn’s moon Titan’s atmosphere to SAR (synthetic aperture radar) and visible/IR data acquired from orbit. A balloon in Titan’s atmosphere must be able to localize itself autonomously both globally and with respect to local terrain. The orbital data is used to provide the balloon imagery with global context.
Due to the highly dissimilar appearance of imagery from the different types of sensors under consideration (radar, IR, visible), traditional image matching techniques based on pixel similarity do not work. Technology pioneered by the medical imaging community has been adapted to match across sensor modalities. These techniques are driven by information content rather than appearance. While imagery of Titan’s surface taken from a visible imager may appear very different from SAR imagery, there is statistical/information theoretic similarity.
The work is novel in applying mutual information (MI) to orbital vs. aerial data. There are unique challenges in this setting. Image offsets are much higher than in medical imaging, there is local distortion due to 3D terrain relief,
and the fields of regard from orbit and from the air are quite different.
Because of the large differences in image scale between an orbiter at hundreds of kilometers above the surface and a balloon at a few kilometers altitude, it is necessary to match mosaics from the balloon to single-frame orbital images. In addition to localizing the balloon, this implies the ability to generate high-resolution global maps of the surface that are correctly geo-referenced.