A sliding window method considering image shadow to detect Mars rock from MRO HiRISE datasets

Abstract

Rock distribution is a crucial factor in landing site selection for Mars exploration. Typically, rocks in flat Martian terrains are characterized by clear boundaries and distinct shadows. We developed a new method (named SSW-ROCK) for rock detection from HiRISE images using the shadow (S) and sliding window technique (SW). SSW-ROCK uses shadows to define the minimum bounding rectangle in the direction of illumination, establishing an initial sliding window based on this rectangle. The window is then slid to the termination position according to the predefined conditions. The rock size can be obtained by fitting the ellipse with the positions of the initial and termination windows. The rock height is estimated using the shadow length along the illumination direction. We used five HiRISE images of Mars between 65° N −70° N for rock detection and detected 532,284 rocks with maximum diameters >1.5 m. We selected accuracy assessment areas in each of the five images and extracted the rocks manually. The SSW-ROCK results were assessed for accuracy using the manual results as a benchmark. In the assessment, we proposed two evaluation metrics, PS and PM: PS measures the proportion of SSW-ROCK results with center points within the range of manual results, while PM measures the proportion of manual results with center points within the range of SSW-ROCK results. Accuracy assessments in five selected areas showed that the mean for both PS and PM exceeded 77 %. Additionally, the dimensions detected by the SSW-ROCK method for known Mars landers closely match their actual sizes. These experiments demonstrate that the SSW-ROCK method is effective for rock detection in flat Martian terrains.