The automation of unmanned aerial vehicles is an active research area. Among different techniques for Unmanned Aerial Vehicle (UAV) automation, obstacle detection is gaining attention to ensure UAV safety flights. This paper presents an algorithm that can be embedded on an UAV to detect forward obstacles. This algorithm is based on a fast and accurate object detection model combining CenterNet and MobileNetV2. This model is pre-trained on the Microsoft COCO 2017 dataset and allowed us to identify obstacle zone in images from the UAV’s front camera. After obstacle detection, this algorithm proposes new directions to be followed by the UAV. These directions are left, right, or up to avoid obstacles. The proposed algorithm detects obstacles and proposes new directions to avoid obstacles by only using a frontal camera. It is sized to be embedded on Raspberry Pi 4b and runs in about 800 ms. It may be combined with the UAV’s coordinates and mission path to regenerate a new mission path in an unknown environment.
Forward Obstacle Detection by Unmanned Aerial Vehicles
Lect.Notes Social.Inform.
Pan-African Artificial Intelligence and Smart Systems Conference ; 2022 ; Dakar, Senegal November 02, 2022 - November 04, 2022
2023-02-21
14 pages
Article/Chapter (Book)
Electronic Resource
English
MOVING OBSTACLE AVOIDANCE FOR UNMANNED AERIAL VEHICLES
British Library Conference Proceedings | 2013
|MONOCULAR VISION-BASED OBSTACLE DETECTION/AVOIDANCE FOR UNMANNED AERIAL VEHICLES
British Library Conference Proceedings | 2016
|CONTROLLING UNMANNED AERIAL VEHICLES TO AVOID OBSTACLE COLLISION
European Patent Office | 2017
|Controlling unmanned aerial vehicles to avoid obstacle collision
European Patent Office | 2020
|Vision-based runway obstacle detection scheme for unmanned combat aerial vehicles
British Library Online Contents | 2007
|