首页|基于多关键点检测加权融合的无人机相对位姿估计算法

基于多关键点检测加权融合的无人机相对位姿估计算法

扫码查看
原文链接
  • 国家科技期刊平台
  • NETL
  • NSTL
  • 万方数据
针对无人机降落阶段中无人船受水面波浪影响导致图像产生运动模糊以及获取无人机相对位姿精度低且鲁棒性差的问题，提出一种基于多模型关键点加权融合的6D目标位姿估计算法，以提高位姿估计的精度和鲁棒性。首先，基于无人船陀螺仪得到的运动信息设计帧间抖动模型，通过还原图像信息达到降低图像噪声的目的;然后，设计一种多模型的级联回归特征提取算法，通过多模型检测舰载视觉系统获取的图像，以增强特征空间的多样性;同时，将检测过程中关键点定位形状增量集作为融合权重对模型进行加权融合，以提高特征空间的鲁棒性;紧接着，利用EPnP(Efficient perspective-n-point)计算关键点相机坐标系坐标，将PnP(Perspective-n-point)问题转化为ICP(Iterative closest point)问题;最终，基于关键点解集的离散度为关键点赋权，使用ICP算法求解位姿以削弱深度信息对位姿的影响。仿真结果表明，该算法能够建立一个精度更高的特征空间，使得位姿解算时特征映射的损失降低，最终提高位姿解算的精度。
Relative Pose Estimation Algorithm for Unmanned Aerial Vehicles Based on Weighted Fusion of Multiple Keypoint Detection
A 6D target pose estimation algorithm based on multiple models keypoints weighted fusion is proposed to address the issue of low accuracy and poor robustness in obtaining the relative pose of unmanned aerial vehicles due to motion blur caused by the influence of water surface waves on the image during the landing phase of un-manned aerial vehicles.This algorithm aims to improve the accuracy and robustness of pose estimation.Firstly,based on the motion information obtained from the unmanned ship gyroscope,an inter frame jitter model is de-signed to reduce image noise by restoring image information.Then,a cascaded regression feature extraction al-gorithm with multiple models is designed to detect images obtained by the shipborne visual system through mul-tiple models,in order to enhance the diversity of the feature space;at the same time,the incremental set of keypo-int localization shapes during the detection process is used as the fusion weight to weight and fuse the model,in or-der to improve the robustness of the feature space.This paper uses efficient perspective-n-point(EPnP)to calcu-late the coordinates of the camera coordinate system for keypoints,and transforms the perspective-n-point(PnP)problem into an iterative closest point(ICP)problem.Finally,based on the dispersion of the keypoints solution set,weights are assigned to keypoints,and the ICP algorithm is used to mitigate the influence of depth information on the pose estimation.The simulation results show that this algorithm can establish a more accurate feature space,re-duce the loss of feature mapping during pose estimation,and ultimately improve the accuracy of pose estimation.

Assisting unmanned aerial vehicles landingshipborne visual system6D pose estimationweighted fu-sionkeypoint detectioncascading feature extraction

葛泉波、李凯、张兴国

展开 >

南京信息工程大学自动化学院 南京 210044

江苏大数据分析与智能系统省高校重点实验室 南京 210044

大气环境与装备技术协同创新中心 南京 210044

中国飞行试验研究院 西安 710089

展开 >

辅助无人机降落 舰载视觉系统 6D位姿估计 加权融合 关键点检测 级联特征提取

国家自然科学基金江苏高校"青蓝工程"

62033010R2023Q07

2024

10.16383/j.aas.c230297

自动化学报
中国自动化学会 中国科学院自动化研究所

自动化学报

CSTPCD北大核心
影响因子：1.762
ISSN：0254-4156
年,卷(期)：2024.50(7)