Crater detection for autonomous landing on asteroids

Abstract

We describe a visual positioning system for use by a spacecraft to choose a landing site, while orbiting an asteroid. The spacecraft pose is refined using landmarks, such as craters, observed by a visual sensor. The craters, which have an elliptical shape, are detected using a multi-scale method based on voting, and tensors as a representation. We propose a new robust method to infer curvature estimation from noisy sparse data. This method is applied on edge images in order to obtain the oriented normals of the edge curves. Using this information, a dense saliency map corresponding to the position and shape of the craters is computed. The detected craters in the image are matched with the craters projected from a 3D model, and the best transformation between these two sets is obtained. This system has been tested with both real images of Phobos and a synthetic model.