Mars Science Laboratory (MSL) surface operations require precise and accurate knowledge of rover position. A key means of establishing/verifying position is to match ortho-rectified mosaics from the MSL onboard cameras to orbital data. Manual localization by matching mosaics to HiRISE imagery can be laborious and somewhat subjective. Ortho-rectified mosaics and orbital images differ dramatically in appearance, due to the extreme viewpoint change as well as occlusions (i.e. objects must be in line of sight) in the mosaics. A straightforward intensity-based matcher using correlation or local feature descriptors cannot cope with this difference. Instead, an information theoretic matcher was used that measures the mutual information between the mosaic and trial positions in the HiRISE imagery.
This work leverages techniques developed for matching orbital to aerial imagery for a hypothetical Titan balloon mission using the Cassini spacecraft as the reference. In this application, HiRISE takes the place of the Cassini SAR (synthetic aperture radar), and rover-based mosaics take the place of images from the balloon. It is assumed that scale and orientation are well known for both the HiRISE imagery and the mosaics. The former comes from SPICE data and the latter from the known imaging geometry of the MSL cameras combined with the onboard INS (inertial navigation system) state. The remaining uncertainty is in translation, expressed as a lateral and vertical shift from the nominal odometry derived position of the rover in the HiRISE image. While the rover itself is not visible (and rarely present) in the HiRISE image, an ortho-rectified mosaic centered on the rover provides enough context to infer the rover position by matching to the orbital data.
The basic information theoretic matching technique employed in this work is well established in theory, and has even been commercialized (in a limited version) in medical imaging applications. However, application of these techniques to the type of imagery under consideration is novel. Further, the notion of automatic rover localization against orbital imagery is novel.
This work, if matured and deployed, will dramatically reduce the man-hours required for rover localization, and reduce the need for expert oversight in this process. Further, it provides concrete metrics on match accuracy instead of a subjective evaluation.