點雲匹配和ICP演算法概述

查志強發表於2016-08-31

【原文:http://www.cnblogs.com/yhlx125/p/4955337.html

Iterative Closest Point (ICP[1][2][3] is an algorithm employed to minimize the difference between two clouds of points.

點雲匹配分類法(1)

•全域性匹配演算法 Globe
•區域性匹配演算法Local

Salvi, J. (2007). "A review of recent range image registration methods with accuracy evaluation." Image and Vision Computing 25(5): 578-596.
Mellado, N. and D. Aiger (2014). "SUPER 4PCS Fast Global Point cloud Registration via Smart Indexing."

點雲匹配分類法(2)

•基於點的匹配
•基於特徵的匹配
•點特徵
•VPF
•FHPF
•…
•基於線特徵
•"Algorithms for Matching 3D Line Sets."
•"Line segment-based approach for accuracy assessment of MLS point clouds in urban areas.“
•Poreba, M. and F. Goulette (2015). "A robust linear feature-based procedure for automated registration of point clouds." Sensors (Basel) 15(1): 1435-1457.

Coarse to fine registration粗-精過程

粗配的目的:提供剛體變換初始估計

Salvi, J., et al. (2007). 

改進ICP演算法

Besl, P. J. and N. D. Mckay (1992). "A Method for Registration of 3-D Shapes." IEEE Transactions on Pattern Analysis and Machine Intelligence 14(2): 239-256.
Siegwart, R., et al. (2015). "A Review of Point Cloud Registration Algorithms for Mobile Robotics." Foundations and Trends in Robotics.

•加快搜尋效率
•K-D樹
•Voronoi圖
•不同的距離量測方式
•點到點
•點到線 PLICP
•Censi, A. (2008). "An ICP variant using a point-to-line metric." IEEE International Conference on Robotics & Automation. IEEE,: 19-25.
•CSM(Canonical Scan Matcher)原始碼     http://censi.mit.edu/software/csm/
•點到面
•Low, K.-L. (2004).   

ICP演算法求解

•Closed Form
•SVD
•Unit Quaternions單位四元數
•The ICP error function minimization via orthonormal matrices
•Dual Quaternions
•數值解法
•LM演算法 (Levenberg-Marquardt algorithm)
•Jerbić, B., et al. (2015). "Robot Assisted 3D Point Cloud Object Registration." Procedia Engineering 100: 847-852.
•點到面線性最小二乘法
•Low, K.-L. (2004). "Linear Least-Squares Optimization for Point-to-Plane ICP Surface Registration."

問題

•觀測誤差
•部分重疊
•離群點Outlier、噪聲(經常是錯誤點或者異常點)
•不滿足一一對應的條件

解決方法

•剔除 Rejection
•PCL類庫中採用
•權重方法
•穩健方法

Bergström, P. and O. Edlund (2014). "Robust registration of point sets using iteratively reweighted least squares."
H. Pottmann, S. Leopoldseder, and M. Hofer. Simultaneous registration of multiple views of a 3D object. ISPRS Archives 34/3A (2002), 265-270.
Andreas Nüchter(2008).3D Robotic Mapping-The Simultaneous Localization and Mapping Problem with Six Degrees of Freedom


標準ICP

標準ICP演算法是最早提出的基於點-點距離的演算法,另外一種是基於點-面的演算法,由chen提出,好多文獻所說的惡Chen's Method。

標準的ICP演算法需要粗配,滿足距離足夠近這一條件之後才能進行精確配準。

IDC

The idc algorithm does a point-to-point correspondence for calculating the scan alignment. The correspondence problem is solved by two heuristics: the closest point rule and the matching range rule. Furthermore, a formula is provided for calculating an error covariance matrix of the scan matching

穩健ICP

由於Outliner的存在,即觀測誤差和離群點存在,以及部分重疊問題,粗配之後的資料再進行精配的過程中仍然存在不穩健的問題(Robust問題),因此提出了穩健ICP方法。如SICP,IRLSICP

MBICP

...

一般的ICP演算法(上述的)是區域性優化演算法,還存在全域性優化的問題,即不需要單獨粗配,直接一步到位。很多的ICP演算法都是穩健的方法,但是並不是全域性的優化方法。全域性的方法有Super4PCS、三點Ransac等。

http://www.mathworks.com/matlabcentral/fileexchange/12627-iterative-closest-point-method

http://www.mathworks.com/matlabcentral/fileexchange/27804-iterative-closest-point

http://projects.asl.ethz.ch/datasets/doku.php?id=laserregistration:laserregistration


相關文章