基于增量式路标表观学习的移动机器人定位  

Approach to mobile robot localization based on incremental landmark appearance learning

作　　者：吴华[1] 秦世引[1] 

机构地区：[1]北京航空航天大学自动化科学与电气工程学院,北京100191

出　　处：《北京航空航天大学学报》2010年第6期708-712,共5页Journal of Beijing University of Aeronautics and Astronautics

基　　金：国家高技术研究发展(863)计划资助项目(2006AA04Z207);教育部博士点基金资助项目(20060006018);中澳国际合作资助项目(2007DFA11530);国家自然科学基金资助项目(60875072)

摘　　要：在全景图像中,机器人参考定位的路标表观受到畸变、观察视角、路标尺度变化以及环境亮度的影响,致使基于全景视觉的机器人自主定位存在着许多难点有待解决.提出增量式的路标表观学习方法,准确估计路标的表观变化,并为基于粒子滤波的机器人定位过程提供准确的观测信息.增量式路标学习过程利用增量式概率主元分析为理论工具,将不同视角的路标表观的主元特征表示成不断自主更新的特征基底,为计算观测量与路标真实表观的相似度提供了实现途径和理论依据.该学习算法能够被集成到带有重采样的贯序权值采样粒子滤波算法过程中,实现了全景视觉机器人的精确自主定位.实验结果表明:该算法的定位误差小,计算量小,执行效率高,对全景图像中的各类干扰均不敏感.Robot localization with omnidirectional vision becomes much difficult as the landmark appearances change dramatically in the omnidirectional image.An approach to mobile robot localization with omnidirectional vision was proposed,which used incremental landmark appearance learning to deal with changes of the landmark appearances and to provide observation information for estimating the robot pose under a particle filtering framework.Incremental probabilistic principal analysis was employed to solve the incremental landmark appearance learning problem,where the landmark appearances viewed from different angles were adopted into the learning model.The proposed method can estimate the robot pose accurately since it takes the advantages of particle filtering by means of sequential importance sampling with resampling.The experimental results demonstrate satisfactory performance with low localization error,little computation burden,efficiency,robustness to various interference in omnidirectional vision.

关 键 词：同步定位与地图构建 计算机视觉 主元分析 奇异值分解 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置]