Non-Cooperative Vision-Based Target Detection and Tracking

The theme of this dissertation is the development of image processing algorithms that enhances the capabilities of the space navigation camera build at the Technical University of Denmark (DTU). The main focus of this dissertation is improvements of the scientific yield of the Planetary Instrument for X-ray Lithochemistry (PIXL) onboard the NASA Mars 2020 rover, Perseverance. The integrated camera system of PIXL is the first to use active light sources directly on an extraterrestrial planetary surface to generate multispectral data. To create multispectral data from the images collected by the monochrome camera of PIXL, a novel radiometric correction method was developed. The method is verified using calibrated spectral data from the the Three Axis N-sample Automated Goniometer for Evaluation Reflectance (TANAGER). The tests show a mean error of 0.4% using an Engineering Model (EM) of the camera system. An evaluation of the radiometric correction method applied on data acquired in-flight show performances within 0.25%.

To analyze the multispectral data, different data products and methods of analyzing, clustering, and classifying were developed. These data products and methods have either been implemented or are in the process of being implemented into NASA’s geoscience visualization and analysis tool, PIXLISE. The multispectral capabilities and analyses of the PIXL multispectral data have proven very valuable for extrapolating the detailed mineral and crystallographic information derived from PIXL’s X-ray spectroscopy system. PIXL is a first-of-its-kind instrument. This has meant that operational procedures have rapidly been developed to optimize the data quality. An in-flight implemented method for acquiring High Dynamic Range (HDR) images with PIXL, which extended the region within the image available for multispectral analysis by 39.4%, is described. Furthermore, a procedure for removing the artifacts of an external light source (the Sun) in the multispectral data was developed to allow higher-quality data to be captured during daytime operation.

Topographical features of a target can have a significant impact both on the data quality and on the co-registration of data. Therefore, knowledge and correction of the topography of a target is essential. A method for obtaining high-resolution topography maps using the PIXL camera system as a simulated stereo-camera by capturing images from different positions is described. This work has led to an update of how topographic maps are generated by PIXL ground tools.
Another objective of this dissertation is the development and implementation of a method for non-cooperative mapping between a 3D vertex model and an image of a target spacecraft. Non-cooperative navigation is essential in the future of space operation, as it provides a method for handling objects that were not designed for cooperative navigation. Non-cooperative navigation can, for instance, help in the process of cleaning up the rising amount of space debris, as this can potentially be approached and removed safely using a designated spacecraft. The described method has been implemented in the programming language C and correctly maps a 3D model to 373 test images, with an average runtime
of 51.5 ms when scaled to a 96 MHz processor.