UAV-Pose: A Dual Capture Network Algorithm for Low Altitude UAV Attitude Detection and Tracking

This paper presents a low-altitude unmanned aerial vehicle (UAV) attitude detection and tracking algorithm, named UAV-Pose. In the context of low-altitude UAV countermeasure tasks, precise attitude detection and tracking are crucial for achieving laser-guided precision strikes. To meet the varying r...

Full description

Saved in:
Bibliographic Details
Published in:IEEE access 2023, Vol.11, p.129144-129155
Main Authors: You, Jiang, Ye, Zixun, Gu, Jingliang, Pu, Juntao
Format: Article
Language:English
Subjects:
Algorithms
Altitude
Atmospheric turbulence
Attitudes
Autonomous aerial vehicles
Convolutional neural networks
Data augmentation
Datasets
Drones
Feature extraction
Frames per second
Low altitude
maneuvering target
Radar tracking
Real-time systems
real-time tracking
Target detection
Target tracking
Task analysis
Tracking
UAV countermeasures
UAV pose detection
Unmanned aerial vehicles
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:This paper presents a low-altitude unmanned aerial vehicle (UAV) attitude detection and tracking algorithm, named UAV-Pose. In the context of low-altitude UAV countermeasure tasks, precise attitude detection and tracking are crucial for achieving laser-guided precision strikes. To meet the varying requirements during the tracking stages, this study designs two capture networks with different resolutions. Firstly, a lightweight bottleneck structure, GhostNeck, is introduced to accelerate detection speed. Secondly, a significant improvement in detection accuracy is achieved by integrating an attention mechanism and SimCC loss. Additionally, a data augmentation method is proposed to adapt to attitude detection under atmospheric turbulence. A self-collected dataset, named UAV-ADT (UAV Attitude Detection and Tracking), is constructed for training and evaluating the target detection algorithm. The algorithm is deployed using the TensorRT tool and tested on the UAV-ADT dataset, demonstrating a detection speed of 300 frames per second (FPS) with a map75 reaching 97.8% and a PCK (Percentage of Correct Keypoints) metric reaching 99.3%. Real-world field experiments further validate the accurate detection and continuous tracking of UAV attitudes, providing essential support for counter-UAV operations.
ISSN:2169-3536
2169-3536
DOI:10.1109/ACCESS.2023.3333394