A new approach of initial localization for planetary lander with the bag of visual words model

Precise localization of a planetary lander during descent phase is crucial for ensuring a safe landing. Terrain Relative Navigation (TRN) is a critical technology for achieving precise localization in hazardous terrains. Initial localization which aligns the descent image from lander camera with the reference map, is essential for TRN to generate precise localization results. This paper introduces a novel initial localization method utilizing the Bag of Visual Words (BoVW) model to pinpoint the descent image on the global map. This method enhances localization efficiency by narrowing the search range for image-map matching. This approach was validated with both synthesized and actual descent image data from the Mars 2020 mission.