基于图像配准的栅格地图拼接方法  

Grid Map Merging Approach Based on Image Registration

作　　者：祝继华[1] 周颐[2] 王晓春[1] 邗汶锌 马亮[1] 

机构地区：[1]西安交通大学软件学院,西安710049 [2]西安交通大学广电中心,西安710049

出　　处：《自动化学报》2015年第2期285-294,共10页Acta Automatica Sinica

基　　金：国家自然科学基金(61203326;u1261111);中国博士后科学基金(2012M512004;2013T60878);陕西省自然科学基金(2014JM8342)资助~~

摘　　要：栅格地图拼接是多移动机器人协同创建环境地图中的一项关键技术.本文提出一种图像配准意义下的栅格地图拼接方法.该方法将栅格地图拼接问题视为图像配准问题,建立相应的目标函数,并给出局部收敛的迭代最近点算法求解该目标函数.为获得最优的拼接结果,该方法从待拼接的地图中提取局部不变特征,并借助随机抽样一致性算法分析初始拼接参数,以作为迭代最近点算法的初值.最后,提出了拼接参数已知时的栅格地图融合规则.实验结果表明,该方法能可靠地实现栅格地图拼接,且具有精度高和速度快的优点.Grid map merging is an important and fundamental issue in cooperative mapping for multi-robot systems. To address this issue, this paper proposes a novel approach to merging grid maps based on image registration. It first turns the map merging problem into the problem of image registration and then presents the corresponding objective function. This function can be solved by the proposed ICP algorithm, which is locally convergent. To obtain the global minimum, SIFT （scale-invariant feature transform） are extracted from the grid maps and are adopted to estimate good initial parameters by the RANSAC algorithm. Furthermore, a rule is also presented to fuse two grid maps with merging parameters. Experimental results demonstrate that the proposed approach can achieve grid map merging with good robustness, accuracy and efficiency.

关 键 词：多移动机器人系统 栅格地图拼接 图像配准 迭代最近点算法 尺度不变特征转换 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP242[自动化与计算机技术—计算机科学与技术]