Karlsson, Samuel (Luleå University of Technology), Kanellakis, Christoforos (Luleå University of Technology), Sharif Mansouri, Sina (Luleå University of Technology), Nikolakopoulos, George (Luleå University of Technology, Sweden)

Monocular Vision-Based Obstacle Avoidance Scheme for Micro Aerial Vehicle Navigation

Scheduled for presentation during the Regular Session "See-and-avoid Systems" (FrC2), Friday, June 18, 2021, 14:00−14:20, Kozani

2021 International Conference on Unmanned Aircraft Systems (ICUAS), June 15-18, 2021, Athens, Greece

This information is tentative and subject to change. Compiled on April 23, 2024

Abstract

One of the challenges in deploying Micro Aerial Vehicless (MAVs) in unknown environments is the need of securing for collision-free paths with static and dynamic obstacles. This article proposes a monocular vision-based reactive planner for MAVs obstacle avoidance. The avoidance scheme is structured around a Convolution Neural Network (CNN) for object detection and classification (You Only Lock Once (YOLO)), used to identify the bounding box of the objects of interest in the image plane. Moreover, the YOLO is combined with a Kalman filter to robustify the object tracking, in case of losing the boundary boxes, by estimating their position and providing a fixed rate estimation. Since MAVs are fast and agile platforms, the object tracking should be performed in real-time for the collision avoidance. By processing the information of the bounding boxes with the image field of view and applying trigonometry operations, the pixel coordinates of the object are translated to heading commands, which results to a collision free maneuver. The efficacy of the proposed scheme has been extensively evaluated in the Gazebo simulation environment, as well as in experimental evaluations with a MAV equipped with a monocular camera.