ISS Monocular Depth Estimation Via Vision Transformer

Monocular depth estimation can be used as an energy-efficient and lightweight backup system to the onboard light detection and ranging (LiDAR) instrument. Unfortunately, the monocular depth estimation is an ill-posed problem. Therefore typical methods resort to statistical distributions of image features. In this work, a deep neural network that exploits Vision Transformer in the encoder is trained in a supervised fashion to solve the regression problem. Specifically, the problem considered is monocular depth estimation with images taken at different positions and distances from the International Space Station (ISS) to measure the network performance in a rendezvous maneuver.