Home > Published Issues > 2020 > Volume 9, No. 10, October 2020 >

A UAV Exploration Method by Detecting Multiple Directions with Deep Learning

Duc Viet Bui, Tomohiro Shirakawa, Hiroshi Sato
National Defense Academy of Japan, Department of Computer Science, Japan

Abstract—Recently, autonomous exploration using robots has been researched and developed for different objectives and requirements. The advancement in image processing using deep learning has made some remarkable results in controlling UAV autonomously in the situation without GPS information. However, there are few types of research on autonomous image-based exploration, especially in the situation that requires the ability to recognize and predict multiple directions from images, which is an important key to perform pathfinding in an exploration mission correctly. In this paper, we propose an approach for this problem by applying a supervised-learning method to predict possible directions from images. We introduce a deep learning architecture using the transfer learning technique to evaluate our dataset. The experiment results show the promising capability of the model for handling situations with multiple directions.

Index Terms—UAV, exploration, GPS, multiple directions, monocular camera, deep learning, transfer learning

Cite: Duc Viet Bui, Tomohiro Shirakawa, Hiroshi Sato, "A UAV Exploration Method by Detecting Multiple Directions with Deep Learning," International Journal of Mechanical Engineering and Robotics Research, Vol. 9, No. 10, pp. 1419-1426, October 2020. DOI: 10.18178/ijmerr.9.10.1419-1426

Copyright © 2020 by the authors. This is an open access article distributed under the Creative Commons Attribution License (CC BY-NC-ND 4.0), which permits use, distribution and reproduction in any medium, provided that the article is properly cited, the use is non-commercial and no modifications or adaptations are made.