室内移动机器人目标定位性准确匹配研究

2019-04-04刘俊超陈志军樊小朝闫学勤

现代电子技术 2019年2期

关键词：移动机器人

刘俊超陈志军樊小朝闫学勤

关键词：扫描匹配; 移动机器人; Kinect传感器; 目标定位; 旋转投影统计; 点云匹配

中图分类号： TN953?34; P242 文献标识码： A 文章编号： 1004?373X（2019）02?0159?04

Research on accurate matching of target positioning for indoor mobile robot

LIU Junchao， CHEN Zhijun， FAN Xiaochao， YAN Xueqin

（School of Electrical Engineering， Xinjiang University， Urumqi 830047， China）

Abstract： For the scan matching technology used in the target positioning system of the indoor mobile robot， the traditional iterative closest point algorithm has a strict requirement for the initial position of the to?be registration point cloud， and is difficult to find the correct corresponding point pair. Therefore， a target positioning algorithm is proposed for the mobile robot on the basis of obtaining the 3D environment point cloud by means of the Kinect sensor， querying the corresponding point pair according to the similarities of rotational projection statistics feature descriptors， and performing scan matching. The point cloud map of an object is obtained by means of the Kinect sensor. The point cloud features are extracted by using the feature extraction algorithm. The rotational projection statistics feature descriptors of two to?be registration point clouds are obtained. The corresponding relationship between two descriptors is estimated by comparing the feature similarity degree of two descriptors. The distance difference matrix algorithm is adopted to remove mismatching points and calculate the initial matching parameter. The ICP registration is performed by using the least square iteration method， so as to obtain the final transformation matrix between point clouds and realize target positioning. The experimental results show that the improved algorithm can effectively improve the matching efficiency and registration accuracy of point clouds， and obtain comparatively accurate target positioning information.

Keywords： scan matching; mobile robot; Kinect sensor; target positioning; rotational projection statistics; point cloud matching

0 引言

机器人目标定位是通过对传感器数据进行处理来估算目标位置的技术，是实现机器人自主导航的基础。而点云配准是移动机器人目标定位的核心问题之一。不少学者针对点云配准技术做了不少研究，其中Besl等最先提出的ICP是最普遍和常用的算法，但存在很多缺陷，如对先验位姿信息敏感、收敛速度慢等[1]。因此，为克服这些缺点，许多学者在最近点选取、误差函数、特征点提取方法的求解等方面做了改进[2?5]。但改进后的算法仍存在許多不足之处。如Bergstr?m等提出了基于鲁棒性M估计的迭代最近点算法，但配准精度和稳定性方面常用的准则函数效果不佳[6]。黄源等针对在不同视角下获得的点云数据，提出一种基于特征提取的点云自动配准算法，但该算法复杂度高[7]。

为提高其配准效率和精度，本文提出了一种基于旋转投影统计（Rotational Projection Statistics，ROPS）描述子改进的 ICP算法，以完成室内移动机器人目标定位性的准确匹配。