室内惯性/视觉组合导航地面图像分割算法  

Floor segmentation algorithm for indoor vision/inertial integrated navigation

作　　者：汪剑鸣[1] 王曦[1] 王胜蓓[1] 李士心[2] 冷宇[3] 

机构地区：[1]天津工业大学电子与信息工程学院,天津300160 [2]天津航海仪器研究所,天津300131 [3]苏州出入境检验检疫局,苏州215021

出　　处：《中国惯性技术学报》2011年第5期553-558,共6页Journal of Chinese Inertial Technology

基　　金：天津市应用基础及前沿技术研究计划(项目名称:室内移动机器人视觉导航新方法研究(10JCYBJC26300)

摘　　要：导航技术是机器人实现自主移动的关键技术之一。针对惯性导航创建全局导航地图困难等问题,提出一种新的惯性/视觉组合导航室内全局地图创建方法。规定机器人只能在地面区域中移动,并利用室内俯视图像建立全局地图,提出一种俯视图像地面区域的自动分割算法。首先,利用主元分析算法对图像的局部颜色特征进行降维;其次,利用聚类算法对地面区域进行自动分割;最后,建立了室内俯视图像数据库并对算法的性能进行了验证。由于第四组图像中包含反光区域,算法的分割结果较差,平均正确分辨率为75%。算法在其他各组的平均正确分割率为85%左右。为提高算法的性能,可在应用本算法前利用反光区域检测算法对图像进行预处理。Navigation is a key technology for autonomous robots,which makes them movable in an unknown environment.To tackle the difficulty of building indoor navigation map for inertial navigation systems,a new map building method for inertial/visual navigation is proposed.By limiting robot＇s movement within the floor areas,the global navigation map is generated from a bird-view image.An algorithm of automatic floor segmentation is proposed,which employs principal component analysis to implement dimension reduction for local color features and adopts clustering analysis to realize floor segmentation automatically.Finally,an indoor bird-view image database is built to evaluate the algorithm.The algorithm gets the worst performance,75% averaged accurate segmentation rate,on the fourth group images because illumination reflection is found in the images.Average accurate segmentation rates on other groups are around 85%.Thus,the preprocessing algorithms,such as illumination refection detection,can help to improve the performance of the algorithm.

关 键 词：室内移动机器人 组合导航 图像分割 聚类分析 

分 类 号：U666.1[交通运输工程—船舶及航道工程]