UAV-Pose: A Dual Capture Network Algorithm for Low Altitude UAV Attitude Detection and Tracking
You Jiang, Zixun Ye, Jingliang Gu, Juntao Pu
Abstract
This paper proposes a low-altitude unmanned aerial vehicle (UAV) pose detection and tracking algorithm, UAV-Pose. In counter-UAV missions, accurate pose detection and tracking are of paramount importance for achieving precision laser targeting. To address different stages of tracking, two capture networks with distinct resolutions are designed. Firstly, a lightweight bottleneck structure, GhostNeck, is introduced to accelerate the detection speed. Secondly, a combination of attention mechanisms and the SimCC loss significantly enhances detection accuracy. A data augmentation method is also proposed to adapt to pose detection under atmospheric turbulence. Finally, the algorithm is deployed using TensorRT and tested on a self-built dataset. Experimental results demonstrate that the proposed approach achieves a performance of 300 FPS on the NVIDIA Jetson NX. Accurate detection and continuous tracking of UAV pose are achieved in real-world field experiments, providing support for countermeasure operations.