Object Recognition Algorithms for the Didymos Binary System

Optical-based navigation in a binary system such as the Didymos one poses new challenges in terms of image processing capabilities, in particular for what concerns the recognition between the primary and secondary bodies. In this work, the baseline object recognition algorithm used in the Milani mission to distinguish between Didymos and Dimorphos is evaluated against alternative image processing pipelines which use convolutional pooling architectures and machine learning approaches. The tasks of the proposed alternatives is to detect the secondary in the image and to define a bounding box around it. It is shown that these algorithms are capable of robustly predicting the presence of the secondary albeit performing poorly at predicting the components of the bounding box, which is a task that is performed quite robustly by the baseline algorithm. A new paradigm is therefore proposed which merges the strengths of both approaches into a unique pipeline that could be implemented on-board Milani.