点云配准(Registration)算法——以PCL为例

Posted 2021-03-12 gdut-gordon

本文为PCL官方教程的Registration模块的中文简介版。

 

An Overview of Pairwise Registration

 

点云配准包括以下步骤：

• from a set of points, identify interest points (i.e., keypoints) that best represent the scene in both datasets;

• at each keypoint, compute a feature descriptor;
• from the set of feature descriptors together with their XYZ positions in the two datasets, estimate a set of correspondences, based on the similarities between features and positions;
• given that the data is assumed to be noisy, not all correspondences are valid, so reject those bad correspondences that contribute negatively to the registration process;
• from the remaining set of good correspondences, estimate a motion transformation.

 

针对上述每一个步骤，PCL的registration模块提供了多种算法进行实现 。

Keypoint

诸如 NARF, SIFT and FAST。

 

Feature descriptors

诸如NARF, FPFH, BRIEF or SIFT。

 

Correspondences Estimation

point matching

• brute force matching,

• kd-tree nearest neighbor search (FLANN),

• searching in the image space of organized data, and

• searching in the index space of organized data.

feature matching

• brute force matching and

• kd-tree nearest neighbor search (FLANN).

 

Corresdondences Rejection

使用RANSAC，或者剪出多余数据。

 

Transformation Estimation

诸如 SVD for motion estimate; - Levenberg-Marquardt with different kernels for motion estimate。

 

算法案例

其中（1）和（2）是point matching，（3）是feature matching。

（1）ICP

ICP的使用SVD求解转换矩阵，其参考文章：

Least-Squares Estimation of Transformation Parameters Between Two Point Patterns

 

（2）NDT

参考论文：

1. The Three-Dimensional Normal-Distributions Transform an Efficient Representation for Registration, Surface Analysis, and Loop Detection. MARTIN MAGNUSSON doctoral dissertation。

2. Line Search Algorithm with Guaranteed Sufficient Decrease. 计算迭代步长。

 

（3）改进版RANSAC

参考论文：

Pose Estimation using Local Structure-Specific Shape and Appearance Context. ICRA 2013.