Computer vision-based navigation systems basically rely on the external environment texture for positioning. This research studies the selection of the computer vision navigation algorithm that suits bet- ter for a known flight environment. Two methods are used to analyze the flight environment images then a comparative study between them is done. While the first method depends on offline spectrum analysis using Fourier transform for the environment images, the second method uses artificial intelligence based on Convolutional Neural Network (CNN) to analyze such images. Usually, in traditional navigation tasks, failures in computer vision navigation systems are solved by depending on mea- surements from other systems that rises the cost, weight, and complexity. This work suggests avoiding the bad matching areas or these where vision failures are expected to take place, by treating them as virtual obstacles and avoiding them using the potential fields method. The previous sug- gestion could result in a stable standalone visual navigation and path tracking system, that is only if no conflict with the mission exists, which is assumed in this research. The two methods are implemented and tested on a path in a simulation environment consisting of a Robot Operating System (ROS), Gazebo simulator, and IRIS drone model. Results show that both methods give indicators to the environment structure, and help in selecting the efficient navigation algorithm. The CNN-based method offers a more detailed description of the environment in form of a meta- map (assistant map) which has been employed effectively to avoid areas over which positioning failures are expected.
Distributed computer and communication networks : control, computation, communications : 25th International conference, DCCN 2022, Moscow, Russia, September 26–29, 2022 : revised selected papers. Cham, 2022. P. 54-66
Language
eng
55 Visitors54 Hits1 Downloads
Факультет инновационных технологий
Spectrum and AI-based analysis for a flight environment and virtual obstacles avoidance using potential field method for path control