一种基于激光测距仪的扫描匹配方法

A scan matching method based on laser range finder

扫描匹配是移动机器人定位和环境构图的关键问题之一,通过计算使相邻扫描重叠度最大的最优变换,从而估算相邻时刻的运动量。ICP算法是目前应用最广泛的扫描匹配算法,针对其对离群扫描点和稀疏扫描点敏感的问题,文中提出一种基于连通格序列的方法对扫描数据进行预处理,从而为扫描匹配提供较为理想的输入条件。此外,针对对应关系中一对多和对应距离极端大的问题,文中通过建立唯一对应,并基于四分位数法计算对应距离的上截断点进行剔除,从而避免建立一对多对应,并减少距离极端大的对应。实验结果表明,文中方法能够剔除扫描中的离群点与稀疏点,并改善ICP算法在实际应用中的扫描匹配表现。

Scan matching is one of key issues in simultaneous localization and mapping,which is used to estimate relative motion by calculating an optimal transformation that maximizes the overlap between two successive scans.ICP is the most widely used scan matching algorithm,but is sensitive to outliers and sparse points in scan.This paper proposes a preprocess method based on continuous grid sequence to reject those outliers and sparse points,and thus provides ideal scans for scan matching.In addition,ICP also suffers from problems of one-to-many correspondences and extremely large distance.This paper solves the problem of one-to-many correspondencesby establishing biunique correspondence, and reject correspondences with extremely large distance by a cutting point based on the quartile method.The experimental result shows that this method effectively reject outliers and sparse points in laser scans,which also has better matching performance in practice.