Feature-Based Vision For Stochastic Motion Tracking Under Partial Occlusion

Abstract

This work is a continuation of the ship-deck landing research presented at VFS Forum 2021, which established a vision-based method to track ship-deck motion before landing when conditions are favorable. This paper makes advancements on two fronts: 1) Demonstrating the effect of velocity feedback on tracking stochastic deck motion, and 2) Developing a new feature-based vision system to detect and track decks in a wide range of challenging environments. The new vision system can detect decks in occlusion as high as 95%. It can also handle a wide range of illumination conditions varying from 20,000 lux of a bright day to 0.01 lux of a dark room. Moreover unlike fiducial-based vision systems that require a custom well designed marker or tag to work, the new vision system can work with generic ship-decks or landing platforms.