Methods have been developed to quantitatively assess rock hazards at candidate landing sites with the aid of images from the HiRISE camera onboard NASA’s Mars Reconnaissance Orbiter. HiRISE is able to resolve rocks as small as 1-m in diameter. Some sites of interest do not have adequate coverage with the highest resolution sensors and there is a need to infer relevant information (like site safety or underlying geomorphology). The proposed approach would make it possible to obtain rock density estimates at a level close to or equal to those obtained from high-resolution sensors where individual rocks are discernable.
The low-resolution data considered here are CTX images, which have a lower resolution than HiRISE images but have a broader span. An important characteristic of CTX and HiRISE images is that they are captured concurrently. Thus, there is a natural pairing between the two data sets.
Bayesian Networks (BNs) are used to graphically model the statistical relationship between rock density estimated from HiRISE images and features extracted from CTX images. Gray Level Co-occurrence Matrix (GLCM) features are used to model the texture in CTX images. The statistical relationship among CTX image features, geomorphology, and rock density is learned by incorporating rock counts from HiRISE images and geomorphic information provided by scientists corresponding to the landing site. The trained BN is then used to infer rock density directly from CTX image features even in the absence of higher resolution images like HiRISE.