基于卷帘快门相机模型的高精度建图与视觉定位
High-Precision Mapping and Visual Positioning Based on Rolling Shutter Camera Model
郭帅威 1肖腾 2邓非1
作者信息
- 1. 武汉大学测绘学院,湖北 武汉,430079
- 2. 湖北工业大学计算机学院,湖北 武汉,430068
- 折叠
摘要
使用卷帘快门相机进行移动测量时,由于逐行曝光的特点,影像上会产生果冻效应.而传统的建图与视觉定位方法大多假设影像是全局快门获取,直接处理卷帘快门影像很难得到高精度的结果.针对以上问题,构建基于插值的卷帘快门相机模型,并将其用于卷帘快门影像的高精度建图与视觉定位.建图时使用运动恢复结构算法,利用卷帘快门相机模型对影像特征点进行坐标插值,获取该点的位姿,继而得到对应的投影矩阵,再进行卷帘快门三角化和卷帘快门光束法平差优化,完成高精度建图.在视觉定位环节,利用PnP(perspective-n-point)算法获取影像定位初值,之后使用卷帘快门影像绝对位姿优化算法进行优化.实验结果表明,与传统方法相比,所提方法的建图与视觉定位精度更高.
Abstract
Images might suffer rolling shutter effect due to the characteristic of row-by-row exposure if a rolling shutter cam-era for movement measurement is used.Traditional methods on mapping and visual positioning typically assume that imag-es are obtained by a global shutter,so,it is difficult to obtain high-precision results when using them to process the rolling shutter images.To solve this problem,we construct a rolling shutter camera model based on interpolation,and use it for high-precision mapping and visual positioning of rolling shut-ter images.Structure from motion(SfM)algorithm is used for mapping,and rolling shutter camera model is used to ob-tain the pose of image feature point by coordinate interpola-tion,and then the corresponding projection matrix is ob-tained,following by performing rolling shutter triangulation and rolling shutter bundle adjustment to complete high-preci-sion mapping.In visual positioning,perspective-n-point(PnP)algorithm is used to obtain the initial value of image positioning,and then it is optimized by rolling shutter abso-lute pose optimization algorithm.The experimental results prove that the proposed method can achieve higher mapping and positioning precision than that of traditional global shutter methods.
关键词
卷帘快门/高精度建图/视觉定位/运动恢复结构(structure/from/motion,SfM)Key words
rolling shutter/high-precision mapping/visual positioning/structure from motion(SfM)引用本文复制引用
出版年
2024
NETL
NSTL
万方数据