Study of Using Deep Learning Nets for Mark Detection in Space Docking Control Images

There is a problem of understanding of relative position between two spacecrafts in process of docking. Use of images from camera set on spacecraft and observing another one is probably effective solution of this problem. Spacecraft position can be calculated using visual marks and knowledge of spatial configuration of observing spacecraft. Detection of visual marks is important part of this solution. Possibility of using deep networks for increase mark detection quality researched. Results of detection using various net architectures compared to current solution that uses improved Viola-Jones detector.