论文部分内容阅读
针对航天器相对姿态估计问题,提出了一种用于单目视觉成像系统的姿态估计方法.在传统核回归方法的基础上,采用训练数据在姿态空间的相似性对视觉输入(图像特征)空间的核函数进行加权,从而学习得到输入变量(图像特征)与目标变量(姿态)的联合概率分布函数,称为接受函数.对于包含未知姿态航天器的图像,通过求取接受函数在姿态空间的最大值,得到目标航天器的姿态估计值.该方法仅需要训练数据学习模型,较其他基于视觉的方法限制更少.对比实验结果证明了该方法在姿态估计方面的优越性,卫星数据集上的实验结果验证了该方法用于航天器姿态估计的有效性.
In order to solve the problem of relative attitude estimation of spacecraft, a pose estimation method for monocular vision imaging system is proposed. Based on the traditional kernel regression method, the similarity of pose data in training space is used to estimate the pose of vision input (image feature) (Image feature) and the target variable (attitude), which is called the acceptance function.For the image containing the unknown attitude spacecraft, by obtaining the function of the acceptance function in the attitude space Maximum, and get the attitude estimation of the target spacecraft.This method only needs to train the data learning model, which is less restrictive than other vision-based methods.Comparison of the experimental results proves the superiority of this method in attitude estimation, The experimental results verify the effectiveness of this method for spacecraft attitude estimation.