| 摘要: |
| 针对摄像机位姿问题提出了一种加权线性方法,其关键思想是通过加权使经典线性方法的代数误差近似于重投影算法的几何误差,从而达到接近于最大似然估计(Levenberg-Marquardt 简称ML)的精度.通过对经典DLT(direct linear transformation)算法和EPnP 算法使用加权的方法,给出了加权DLT 算法(WDLT)和加权EPnP 算法(WEPnP).大量模拟数据和真实图像实验结果均表明,WDLT 和WEPnP 算法不仅能提高DLT 和EPnP 算法的精度,而且在深度较小的情况下优于Lu 的非线性算法. |
| 关键词: 摄像机姿态 DLT(direct linear transformation) EPnP WDLT(weighted DLT) WEPnP 最大似然估计 |
| DOI：10.3724/SP.J.1001.2011.03916 |
| 分类号: |
| 基金项目:国家自然科学基金(61075038, 60835003) |
|
| Weighted Linear Methods for the Camera Pose Estimation |
|
YANG Sen, WU Fu-Chao
|
|
National Laboratory of Pattern Recognition, Institute of Automation, The Chinese Academy of Sciences, Beijing 100190, China
|
| Abstract: |
| This paper presents a novel weighted linear method for the camera pose estimation. The key idea of this method is to replace the algebraic error in the classic linear method with the weighted algebraic error that closes the geometric error. The method provides a linear solution whose accuracy is close to the accuracy of an ML estimation. Based on the DLT (direct linear transformation) algorithm and EPnP algorithm, the weighted DLT (WDLT) and weighted EPnP (WEPnP) algorithms are obtained by using the weighted linear technique. Experimental results with simulative data and real images show that the WDLT and WEPnP algorithms remarkably outperform the DLT and EPnP algorithms and in the case of small depth ratio, both of them outperform the Lu’s nonlinear algorithm. |
| Key words: camera pose DLT (direct linear transformation) EPnP WDLT (weighted DLT) WEPnP ML estimation |
