Scientific Visualization, 2018, volume 10, number 4, pages 1 - 12, DOI: 10.26583/sv.10.4.01
Segmentation and visualization of obstacles for the enhanced vision system using generative adversarial networks
Authors: V.V. Kniaz1,A,B, S.Yu. Danilov2,A,B, A.N. Bordodymov3,A
A State Res. Institute of Aviation Systems (GosNIIAS), Moscow, Russia
B Moscow Institute of Physics and Technology (MIPT), Russia
1 ORCID: 0000-0003-2912-9986, vl.kniaz@gosniias.ru
2 ORCID: 0000-0003-1346-1685, danilov@gosniias.ru
3 ORCID: 0000-0001-8159-2375, bordodymov@gmail.com
Abstract
An increasing amount of air traffic in general aviation generates a demand to increase the situational awareness of the crew during task intensive stages of the flight such as take-off, landing and taxi. The Head-up Display (HUD) is a prospective device that provides aid for the crew by projection of a flight information over on a display over the cockpit view. Nevertheless, projection of a video sequence on the HUD is challenging due large areas of a high intensity and low contrast of dynamic obstacles (aircraft crossing the flight trajectory and vehicles on a runway).
This paper is focused on the development of an algorithm for detection of dynamic obstacles in the air and on the runway. To achieve this task the algorithm leverages an original deep convolutional neural network ObstacleGAN.The developed algorithm performs parallel data processing and can be implemented using the modern integrated modular avionics. The numerical complexity of the algorithm allows to implement it for processing of enhanced vision system video sequence with resolution of 640?480 pixels using three CPU at 1.8 GHz and one lightweight GPU.
Keywords: aircraft, increasing of a situational awareness, aircraft traffic safety, head-up display, deep convolutional neural networks.
