• CN:11-2187/TH
  • ISSN:0577-6686

机械工程学报 ›› 2018, Vol. 54 ›› Issue (15): 15-20,30.doi: 10.3901/JME.2018.15.015

• 机构学及机器人 • 上一篇    下一篇

扫码分享

一种有效的自主机器人迭代最近点定位算法

杨晶东, 孙磊明, 邵雨婕, 师艳伟   

  1. 上海理工大学光电信息与计算机工程学院 上海 200093
  • 收稿日期:2017-08-17 修回日期:2018-01-17 出版日期:2018-08-05 发布日期:2018-08-05
  • 作者简介:杨晶东,男,1973年出生,博士,副教授。主要研究方向为智能机器人,计算机视觉。E-mail:eerfriend@yeah.net
  • 基金资助:
    国家自然科学基金(61374039)、上海市自然科学基金(15ZR1429100)、沪江基金(C14002)和上海市数据科学重点实验室开放课题(201609060003)资助项目。

Efficient Iterative Closest Point Localisation Algorithm for Autonomous Robots

YANG Jingdong, SUN Leiming, SHAO Yujie, SHI Yanwei   

  1. School of Optical-Electrical and Computer Engineering, University of Shanghai for Science and Technology, Shanghai 200093
  • Received:2017-08-17 Revised:2018-01-17 Online:2018-08-05 Published:2018-08-05

摘要: 迭代最近点算法(Iterative closest point,ICP)在一定程度上可获得较好配准效果,但算法收敛较依赖于输入初始值,容易造成局部最优。正态分布变换(Normal distribution transform,NDT)算法虽精度较高,但需要扫描点数量较多,在长距离导航中会导致较大的转角偏差。提出一种基于滤波ICP自主定位方法(FICP)。构建了基于ICP算法的激光扫描点特征匹配误差模型,采用多种滤波器减少匹配噪声,加快ICP算法收敛速度。推导了滤波后匹配点位姿更新算法,提高全局定位精度。试验表明,相对于传统ICP算法和NDT算法,FICP算法具有较好实时性和定位精度。

关键词: 迭代最近点, 滤波, 全局定位, 扫描匹配

Abstract: The iterative closest point (ICP) algorithm can achieve good performance in the registration of laser scans to some extent. But the convergence depends much on the inputs. It is easy to lead to the local optimal. Normal distribution transform (NDT) algorithm has a higher precision than ICP sometimes, but a large amount of scans are needed for NDT algorithm in the process of matching. The larger angular deviation will happen during long-term autonomous navigation for mobile robots. An improved ICP algorithm is proposed. The error model of the feature matched scans basing on ICP algorithm is derived. The multiple data filters are applied to the features match to reduce the noises of the scans and speed up converge. Also the pose update algorithms are derived to improve the accuracy of global localization after filtering. The experiments show that FICP algorithm, compared with ICP or NDT, has a higher localization accuracy and a better real-time.

Key words: filtering, global localization, iterative closest point, scan matching

中图分类号: