Fast Descriptors and Correspondence Propagation for Robust Global Point Cloud Registration

Abstract

In this paper, we present a robust global approach for point cloud registration from uniformly sampled points. Based on eigenvalues and normals computed from multiple scales, we design fast descriptors to extract local structures of these points. The eigenvalue-based descriptor is effective at finding seed matches with low precision using nearest neighbor search. Generally, recovering the transformation from matches with low precision is rather challenging. Therefore, we introduce a mechanism named correspondence propagation to aggregate each seed match into a set of numerous matches. With these sets of matches, multiple transformations between point clouds are computed. A quality function formulated from distance errors is used to identify the best transformation and fulfill a coarse alignment of the point clouds. Finally, we refine the alignment result with the trimmed iterative closest point algorithm. The proposed approach can be applied to register point clouds with significant or limited overlaps and small or large transformations. More encouragingly, it is rather efficient and very robust to noise. A comparison to traditional descriptor-based methods and other global algorithms demonstrates the fine performance of the proposed approach. We also show its promising application in large-scale reconstruction with the scans of two real scenes. In addition, the proposed approach can be used to register low-resolution point clouds captured by Kinect as well.

DUGMA: Dynamic Uncertainty-Based Gaussian Mixture Alignment

“…We ran all the algorithms on a laptop with Intel Core i7-7820HK processor (quad-core, 8MB cache, up to 4.4GHZ) and NVidia Geforce GTX 1080 with 8GB GDDR5X. To test the accuracy and robustness of our algorithm, our proposed method is compared with relevant recent algorithms from the top journals and conferences: (R, t) ← arg min R,tL CPD [19], GMMREG [15], BDICP [32], GOICP [29], GOGMA [4], 3DMATCH [30] 1 , FDCP [16] 2 . All the code is directly from the authors.…”

“…Similarly, Elbaz et al [9] used a deep neural network auto-encoder to encode local 3D geometric structures instead of traditional descriptors. Lei et al [16] proposed a fast descriptor based on eigenvalues and normals computed from multiple scales to extract the local structure of the point clouds and then recovered the transformation from matches. However, they are sensitive to noisy point clouds.…”

“…Finding the accurate transformation between two noisy rigid point clouds is generally hard: true point-to-point correspondences seldom exist, which limits the accuracy of the methods based on ICP [29,32,23,1,5,12,21]. As for the methods based on local descriptors [16,30,22,25], coarse points tend to cause inaccurate local descriptors to mismatch with each other. Also, the variable density (distant areas have a lower point cloud density) will make them unstable.…”

DeepMapping: Unsupervised Map Estimation From Multiple Point Clouds

“…Pairwise global registration: the global methods [49,4,33,53,31,16] do not rely on the "warm start" and can be performed on point clouds with arbitrary initial poses. Most global methods extract feature descriptors from two point clouds.…”

Effective scaling registration approach by imposing emphasis on scale factor

“…As the variants of ICP algorithm are well known to be susceptible to local convergence, the particle filter [11] or genetic algorithm [12] could be utilized to obtain the desired global minimum. Besides, effective features [13][14][15] can also be extracted and matched for the point sets to be registered so as to provide the initial parameters for registration approaches. These registration approaches may obtain accurate results for the rigid registration, but they may not achieve the scaling registration, which may exist in some practical applications.…”