Estimation of orbital object orientation parameters based on results of space robot observation using machine vision algorithms

Abstract

The problem of estimating the orientation parameters of a spacecraft observed from a space robot is considered, and various approaches to its solution are presented. Results of the study of the accuracy of two approaches to estimating the parameters of the spacecraft orientation from results of a space robot observation using the technical vision methods and convolutional neural networks, are presented. In the first approach, the neural network is used to determine the screen coordinates of the object's special points projections in the image. The second approach is based on the use of a trained convolutional neural network, which directly generates the Euler angles from the observed image.