PTRNet: Global Feature and Local Feature Encoding for Point Cloud Registration

Cuixia Li,Shanshan Yang,Yinghao Li,Li Shi,Yue Liu

doi:10.3390/app12031741

Abstract

Existing end-to-end cloud registration methods are often inefficient and susceptible to noise. We propose an end-to-end point cloud registration network model, Point Transformer for Registration Network (PTRNet), that considers local and global features to improve this behavior. Our model uses point clouds as inputs and applies a Transformer method to extract their global features. Using a K-Nearest Neighbor (K-NN) topology, our method then encodes the local features of a point cloud and integrates them with the global features to obtain the point cloud’s strong global features. Comparative experiments using the ModelNet40 data set show that our method offers better results than other methods, with a mean square error (MSE), root mean square error (RMSE), and mean absolute error (MAE) between the ground truth and predicted values lower than those of competing methods. In the case of multi-object class without noise, the rotation average absolute error of PTRNet is reduced to 1.601 degrees and the translation average absolute error is reduced to 0.005 units. Compared to other recent end-to-end registration methods and traditional point cloud registration methods, the PTRNet method has less error, higher registration accuracy, and better robustness.

Highlights

In recent years, with the increasing maturity of laser technology, a variety of laser systems play an important role in cultural relic restoration, military aerospace, 3D topography measurement, target recognition and other applications
We propose an end-to-end registration model called Point Transformer for Registration Network (PTRNet) that takes both local and global features into account
It is worth noting that the iPCRNET model does not contain the Transformer module and the local feature extraction method adopted by us

Summary

Introduction

With the increasing maturity of laser technology, a variety of laser systems play an important role in cultural relic restoration, military aerospace, 3D topography measurement, target recognition and other applications. Traditional point cloud registration methods are often called optimization-based point cloud registration frameworks Such registration algorithms [1,2,3,4,5,6,7] obtain the optimal transformation matrix by iterating through two phases [8] correspondence search and transformation estimation. The correspondence search finds the corresponding (matching) points between the point clouds Transformation estimation uses these corresponding relations to estimate the transformation matrix. The advantage of these methods is that they do not need training sets, relying on strict data theory to ensure their convergence. They need many complex strategies to overcome noise, outliers, density changes, and partial overlaps, all of which increase the computational cost

Methods

Results

Conclusion