机器人2013，Vol.35Issue(4)：491-499,512,10.DOI:10.3724/SP.J.1218.2013.00491
基于“原型”的机器人开放式室内场所感知实验研究
Experimental Study on Open Interior-Places Perception of Robot Based on "Prototype"
摘要
Abstract
A place-perception experimental platform is proposed on the basis of the prototype based place perception algorithm.In the hardware system,a binocular rig is used as environment perception sensor,and CPU (central processing unit) and GPU (graphic processing unit) coordinated computing unit is used to process multiple tasks,complex logic,intensive data,and so on.The software system is constructed based on Microsoft Robotics Developer Studio platform.Binocular vision service and place perception service are two core services in it.The former is fit for clustered indoor environment,and realizes object recognition and pose estimation simultaneously in wide viewing angle by using RANSAC (RANdom SAmple Consensus),ASIFT (affine scale-invariant feature transform)-GPU and other algorithms.The latter realizes place categorization and region perception and constructs 2D semantic map based on the results of binocular vision and the existing place perception algorithm.The experiments are conducted in a real interior environment in which no artificial label is used.The experiment results show that the robot can robustly perceive open places in its action space online based on the experimental platform,and realizability,effectiveness and practicability of the existing place perception algorithm are verified to a certain extent.关键词
场所分类/开放式室内场所/原型理论/场所感知/GPU/ASIFT/语义地图Key words
place categorization/ open interior place/ prototype theory/ place perception/ GPU (graphic processing unit)/ ASIFT (affine scale-invariant feature transform)/ semantic map分类
信息技术与安全科学引用本文复制引用
朱博,戴先中,李新德,杨伟,陈芳园..基于“原型”的机器人开放式室内场所感知实验研究[J].机器人,2013,35(4):491-499,512,10.基金项目
国家自然科学基金资助项目(60804063,61175091) （60804063,61175091）
江苏省自然科学基金资助项目(BK2010403) （BK2010403）
图像信息处理与智能控制教育部重点实验室开放基金资助项目(200902) （200902）
东南大学优秀青年教师教学、科研资助计划资助项目(3208001203) （3208001203）
东南大学创新基金资助项目(3208000501). （3208000501）
