Efficient Registration of Intraoral Point Clouds Using 3-D Spatial Consistent Grouping Method With Global Optimization

Intraoral 3-D scanning is a vital component of modern oral therapy. Because of the limited field of view (FOV) and degraded image quality of the intraoral scanner, matching the bundle of the intraoral dense point clouds with a low signal-to-noise ratio (SNR) in real time (approximately 25 point clouds per second) is challenging; consequently, the accuracy of the reconstructed full arch gets degraded. This study proposes an innovative matching method for intraoral point clouds with high efficiency, which involves a 3-D spatial consistent grouping (3D-SCG) method in the intraoral scanning scheme. First, the method introduces the Tooth-Net module, which is a multilayer neural network based on the attention mechanism, to achieve high-dimensional geometric features using the corresponding pairs between intraoral point clouds to be matched. Subsequently, outliers are removed based on the geometric features in the outlier removal module, which comprises Score-Net, nonmaximum suppression (NMS), and K-nearest neighbor (KNN) with spectral matching (SM). High confidence by Score-Net and well-distributed correspondences by NMS are converted into scores, and KNN is used to reduce the size of the high-score matrix. The weights generated by the SM module for each corresponding pair are used to calculate the best matching matrix. Four small- and four large-scale high-density intraoral point-cloud datasets are collected to validate the proposed method. Both small- and large-scale intraoral point clouds can be matched in real time. The overall time consumption is reduced by approximately 80% compared with the traditional matching method for intraoral point clouds, which is based on fast point feature histograms (FPFHs) with an iterative closest point (ICP) algorithm; nonetheless, the matching accuracy is comparable. Moreover, the matching failure rate is reduced by 33% during intraoral scanning. The proposed matching method based on 3D-SCG is efficient for intraoral point clouds with a low SNR and high density.