基于合作目标的单目视觉位姿估计  

Monocular Vision Pose Estimation Based on Cooperative Target

作　　者：范路生 张静远[1] 王为如 吴佩璋 FAN Lusheng;ZHANG Jingyuan;WANG Weiru;WU Peizhang(Naval University of Engineering,Wuhan 430000;713 Research Institute of China State Shipbuilding Corporation Limited,Zhengzhou 450015)

机构地区：[1]海军工程大学,武汉430000 [2]中国船舶集团有限公司第七一三研究所,郑州450015

出　　处：《舰船电子工程》2021年第7期80-84,共5页Ship Electronic Engineering

摘　　要：无人物流已经成为全球物流行业的发展趋势,叉车的定位、导航、自主搬运货物等问题亟待解决,视觉位姿估计应用到无人物流行业则是一个新兴热门的研究方向。本文以叉车搬运货物过程中货叉与货叉孔的相对位置、相对姿态为研究对象,基于PnP算法求解两者的相对六维位姿。论文采用Aruco合作标志,先将提取得到角点图像坐标进行线性算法直接线性变换初步得到两者的相对六维位姿,再将其作为初值采用光束法平差对其进行六维位姿优化,构建重投影误差模型函数,采取Gauss-Newton方法进行多次最小二乘拟合迭代,最终得到货叉与货叉孔的高精度相对位姿。Unmanned logistics has become the development trend of global logistics industry.The problems of forklift position⁃ing,navigation and independent cargo handling need to be solved.The application of visual pose estimation to unmanned logistics industry is a new and hot research direction.In this paper,the relative position and relative attitude of fork and fork hole in the pro⁃cess of forklift handling goods are taken as the research object,and the relative six dimensional pose of fork and fork hole is solved based on PnP algorithm.In this paper,the Aruco cooperation mark is used.Firstly,the coordinates of the corner image are extracted and processed by linear algorithm direct linear transformation to obtain the relative six dimensional pose of the two.Then,it is used as the initial value to optimize the six dimensional pose by using the bundle adjustment,and the reprojection error model function is constructed.The Gauss Newton method is used to carry out multiple least square fitting iterations,and the final high precision rela⁃tive pose of fork and fork hole is obtained.

关 键 词：视觉位姿估计 直接线性变换 光束法平差 Gauss-Newton 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]