Evaluation of Methods for Distance Estimation to an Aerial Coupling Device using a Single RGB Image

Assisted or autonomous in-flight coupling of aircraft, including air-to-air refueling and capturing of reusable launch vehicles for spaceflight, requires precise knowledge of the coupling device’s position relative to the trailing aircraft. While cameras and Machine Learning (ML) algorithms excel in bounding box detection, distance estimation from RGB images remains a challenging problem due to the inherent lack of depth information. Against this background, this work investigates eight approaches, including both ML-based and non-ML-based methods, for estimating the distance to a coupling device of known size using a single RGB image. This paper presents, implements, and evaluates each approach, highlighting their accuracy, speed, strengths and limitations. Overall, this work contributes to the advancement of autonomous in-flight coupling by identifying the most effective distance estimation approach, ultimately enhancing the safety and practicality of these critical operations.