Detecting Mars Surface Change Using Landmark Comparison

Current pixel-based methods of automatic image change detection are too computationally expensive and prone to spurious detection to discover meaningful changes in Mars surface features in NASA’s large, continuously expanding database of satellite imagery. Automatically finding these changes is important to planetary geologists as the changes often represent the appearance of new dust devil tracks, dark slope streaks, and meteor impacts. We developed a more efficient method of change detection that compares information about landforms appearing in subsequent images of the same region. By avoiding the direct comparison of pixels, this method reduces computational complexity and is more robust against variations in lighting and camera orientation. We investigated improvements to our methods of selecting regions that stand out from their surroundings and developed a graph representation of these landforms. We then applied graph matching algorithms to detect emerging and disappearing landforms. Finally, we experimented with a RANSAC algorithm to refine this matching under the assumption that the coordinate systems of the compared images are related by an affine transform.