Point Cloud Registration Method Research Articles

Deep Learning Point Cloud Registration based on Distance Features

In this paper, a new method of rigid point cloud registration called Points Registration Learning (PREL) is presented. This algorithm is based on Deep Neural Networks trained by sparse autoencoders and fed with a set of Euclidean and Mahalanobis distance maps. Unlike other reported methods, we do not assume closeness between point clouds or point pairs. This allows registering point clouds with a high degree of displacement or occlusion. PREL algorithm does not require an iterative process, it estimates points distribution non-parametrically and it does not require a finer adjustment using other methods such as Iterative Closest Point (ICP). To evaluate the proposed algorithm, two kinds of point cloud sets were used: one of them corresponds to real scenes acquired with an RGB-D camera and the other set are surface reconstructions. When comparing PREL, ICP and Deep Closest Point (DCP) with Root Mean Square Error (RMSE), using points sets with a high degree of occlusion and displacement, ICP method shows an average RMSE of 98.8, followed by DCP with 32.51 and PREL with 0.75. These results suggest that PREL algorithm can be useful to reconstruct scenes, to scan objects and to register point clouds in any application, given the learning ability of the proposed algorithm.

Point cloud registration method based on SIFT and geometry feature

In order to improve the registration accuracy, the algorithm framework of system is proposed. The object is extracted based on threshold segmentation. The initial corresponding points are acquired based on scale-invariant feature transform (SIFT) operator. Furthermore, the curvature are performed to get rid of the wrong points. With proposed estimation method of overlapped parts, the fine registration is carried out. The experiments are utilized to verify the superiority of proposed method.

Point Cloud Registration Method for Pipeline Workpieces Based on 4PCS and Improved ICP Algorithms

A new point cloud registration method based on 4PCS and improved ICP algorithm is proposed to solve the problem of point cloud registration of laser scanning workpiece position and pose data on industrial pipeline. Firstly, feature points set of point cloud is extracted from normal vector. Then, the rough registration of point cloud is realized by using the registration algorithm based on feature points (4PCS), and the initial recognition of workpiece position and pose is completed. Finally, based on the traditional ICP algorithm, the k-d tree is used to accelerate the search speed of the corresponding point pairs and complete the accurate registration of the point clouds. The proposed algorithm and the 4PCS+ICP algorithm are tested respectively, and the results show that the algorithm is faster, with lower error and fewer iterations.

Point Cloud Registration Method for Pipeline Workpieces Based on RANSAC and Improved ICP Algorithms

Aiming at the registration problem of laser-scanned workpiece point cloud data, a point cloud registration method based on RANSAC algorithm and improved ICP algorithm is proposed. Firstly, feature points are selected according to the variation law of the normal vector in the original point cloud data, and the initial matching point set is obtained by establishing the histogram (FPFH) of feature points. Then a random sampling consensus (RANSAC) algorithm is applied to the initial data matching. At last, the nearest point iterative algorithm (ICP) accelerated by k-d tree is used for accurate matching, and the quaternion method is used to obtain the registration parameters. The new algorithm and PCA+ICP algorithm are tested and the experimental results are compared. The results show that the new algorithm can achieve registration, and improve the speed and accuracy of registration, which provide a reference for similar problems.

Point Cloud Registration Method for Pipeline Workpieces Based on PCA and Improved ICP Algorithms

Aiming at the registration problem of point cloud data obtained by laser scanning workpiece, a new automatic registration algorithm of point cloud based on PCA algorithm and improved ICP algorithm is proposed. Firstly, the feature points are selected according to the change rule of normal vectors in the original point cloud data, and the initial matching point set is obtained by establishing the histogram of feature points (FPFH); then the principal component analysis (PCA) is used to match the initial data; Finally using the k-d tree to accelerate the improved iterative method is closest point (ICP) precise matching, and using quaternion method of registration parameters are obtained. Experiments are carried out on the proposed new algorithm and PCA+ICP algorithm, and the experimental results are compared. The results show that the new algorithm can achieve registration, and improve the speed and accuracy of registration, which provide a reference for similar problems.

Robust Generalized Point Cloud Registration With Orientational Data Based on Expectation Maximization

This paper introduces a robust generalized point cloud (PC) registration method that utilizes not only the positional but also the orientation information associated with each point. The proposed method solves the rigid PC registration problem in a probabilistic manner, which casts the problem into a maximum likelihood (ML) framework. A hybrid mixture model (HMM) is utilized to represent one generalized PC. In the HMM, a von-Mises–Fisher mixture model (FMM) is adopted to model the orientational uncertainty, while a Gaussian mixture model (GMM) is used to represent the positional uncertainty. An expectation–maximization (EM) algorithm is adopted to solve the optimization problem in an iterative manner to find the optimal rotation matrix and the translation vector between two generalized PCs. In both expectation step (E step) and maximization step (M step), orientational information is utilized, which can potentially improve the algorithm’s robustness to noise and outliers. In the E step, the posterior probabilities that represent the degree of point correspondences in two PCs are computed. In the M step, an efficient closed-form solution to a rigid transformation matrix is developed. E and M steps will iterate until certain convergence criteria are satisfied. Extensive experiments under different noise levels and outlier ratios have been carried out on a data set of femur bone computed tomography images. Experimental results show that the proposed method outperforms the state-of-the-art ones in terms of accuracy, robustness, and convergence speed significantly. Note to Practitioners —This paper was motivated by solving the problem of registering two PCs. Most existing approaches generally use only the positional information associated with each point and thus lack robustness to noise and outliers. This paper suggests a new robust method that also adopts the normal vectors associated with each point. The registration problem is cast into a maximum likelihood (ML) problem and solved under the expectation–maximization (EM) framework. Closed-form solutions for estimating parameters in both expectation and maximization steps are provided in this paper. We have demonstrated through extensive experiments that the proposed registration algorithm achieves improved accuracy, robustness to noise and outliers, and faster convergence speed.

Scale‐aware camera localization in 3D LiDAR maps with a monocular visual odometry

AbstractLocalization information is essential for mobile robot systems in navigation tasks. Many visual‐based approaches focus on localizing a robot within prior maps acquired with cameras. It is critical where the Global Positioning System signal is unreliable. In contrast to conventional methods that localize a camera in an image‐based map, we propose a novel approach that localizes a monocular camera within a given three‐dimensional (3D) light detection and ranging (LiDAR) map. We employ visual odometry to reconstruct a semidense set of 3D points from the monocular camera images. These points are continuously matched against the 3D prior LiDAR map by a modified feature‐based point cloud registration method to track a full six‐degree‐of‐freedom camera pose. Since the monocular camera suffers from the scale‐drift problem due to the lack of depth information, the proposed method solves it by adopting updatable scale estimation. Experiments carried out on a publicly large‐scale data set demonstrate that the camera and LiDAR multimodal data matching problem is solved, and the localization accuracy of our method is comparable to state‐of‐the‐art approaches.

An automatic 3D registration method for rock mass point clouds based on plane detection and polygon matching

Point cloud registration is an essential step in the process of 3D reconstruction. Considering that the surface of rock mass is complex and mainly composed of planes, in this paper, we introduce a novel and automatic 3D registration method for rock mass point clouds based on plane detection and polygon matching. Firstly, planes are detected from rock mass point clouds by an efficient tripe-region growing method, and then, the corresponding polygons are calculated by concave hull method. Secondly, PCA-based polygon matching procedure is used for coarse registration. Finally, ICP method is applied to fine registration. The performance of this method was tested in different rock mass point clouds. Compared with the existing methods, the proposed method demonstrates a reliable and stable solution for accurately registering in rock mass scenes.

An Accurate TLS and UAV Image Point Clouds Registration Method for Deformation Detection of Chaotic Hillside Areas

Deformation detection determines the quantified change of a scene’s geometric state, which is of great importance for the mitigation of hazards and property loss from earth observation. Terrestrial laser scanning (TLS) provides an efficient and flexible solution to rapidly capture high precision three-dimensional (3D) point clouds of hillside areas. Most existing methods apply multi-temporal TLS surveys to detect deformations depending on a variety of ground control points (GCPs). However, on the one hand, the deployment of various GCPs is time-consuming and labor-intensive, particularly for difficult terrain areas. On the other hand, in most cases, TLS stations do not form a closed loop, such that cumulative errors cannot be corrected effectively by the existing methods. To overcome these drawbacks, this paper proposes a deformation detection method with limited GCPs based on a novel registration algorithm that accurately registers TLS stations to the UAV (Unmanned Aerial Vehicle) dense image points. First, the proposed method extracts patch primitives from smoothed hillside points, and adjacent TLS scans are pairwise registered by comparing the geometric and topological information of or between patches. Second, a new multi-station adjustment algorithm is proposed, which makes full use of locally closed loops to reach the global optimal registration. Finally, digital elevation models (DEMs, a DEM is a numerical representation of the terrain surface, formed by height points to represent the topography), slope and aspect maps, and vertical sections are generated from multi-temporal TLS surveys to detect and analyze the deformations. Comprehensive experiments demonstrate that the proposed deformation detection method obtains good performance for the hillside areas with limited (few) GCPs.

Frequency domain point cloud registration based on the Fourier transform

Due to the limited measurement range and occlusion of single-line structured light, it is impossible to detect the side data of the whole part. It is proposed that point cloud registration method obtained from multiple rotations of parts in frequency domain by Fourier transform. In the process of point cloud registration, cross-section point cloud data are restored to the corresponding size matrix firstly. Secondly, Fourier transform is carried out to calculate the point cloud data. When calculating the rotation angle, the polar coordinate transformation is carried out at first, and then the cross power spectrum of the two matrices is obtained, so that the rotation and translation matrix of the point cloud can also be obtained. In this process, considering point cloud noise existence, the Sinc function is approximately replaced by the non-noise inverse Fourier transform of cross power spectrum, so that the noise has no influence on the determination of registration parameters in frequency domain registration. The registration accuracy of point cloud is checked by high precision rotation and multiple measurements of mobile platform. Finally, the rotation matrix and translation values are obtained.

A Hierarchical Skull Point Cloud Registration Method

Skull registration is one of the important steps in craniofacial reconstruction, and its registration accuracy and efficiency have an important impact on the reconstruction results. To solve the problem of low accuracy and efficiency of existing skull registration methods, a hierarchical skull point cloud registration method is proposed in this paper. The whole registration process is divided into a rough registration stage and a fine registration stage. Firstly, feature points are extracted from the pre-processed skull point cloud model, and a local coordinate reference system is established according to the feature points and their neighbor points. The improved spin image is used to construct the local feature descriptor. The feature matching is carried out according to the nearest neighbor algorithm, and the k-means algorithm is used to eliminate the mismatching points to achieve skull rough registration. Then, based on rough registration, we use an improved ICP algorithm to achieve fine registration of the skull. In this process, we use random sampling to reduce the search scale of points and add geometric feature constraints to further eliminate mismatched points. Finally, the whole registration algorithm is applied to the skull point cloud data to verify. The experimental results show that, compared with other methods, the registration effect and efficiency of the proposed method are superior to those of other methods. In order to verify the universality of the method, we also use a common data set for verification. Experiments show that the method is also very effective.

基于对偶四元数构建的直线基元点云拼接方法

基于对偶四元数构建的直线基元点云拼接方法

GP-ICP: Ground Plane ICP for Mobile Robots

In this paper, we propose a robust point cloud registration method for ground vehicles. Given the vast developments in the field of autonomous vehicles, the use of point cloud data has increased. The simultaneous localization and mapping (SLAM) algorithm is typically used to generate sophisticated point cloud maps. In the SLAM algorithm, the quality of the map depends on the performance of loop closure algorithms. The iterative closest point (ICP) algorithm is widely used for loop closure of the point cloud. However, the ICP algorithm might not work well for ground vehicles because it was originally developed for 3D reconstruction in computer vision field. Therefore, this paper proposes a method to find robust matching correspondences in the ICP algorithm on ground vehicle conditions. The performance of the proposed method is compared with other conventional methods by using KITTI open datasets. The source code is publicly released on the Github website.

Registration method of point clouds using improved digital image correlation coefficient

Three-dimensional (3-D) modeling is widely applied in the fields of measurement engineering and manufacturing. Registration of point clouds is a crucial step in 3-D model reconstruction. We present a registration method using improved digital image correlation (DIC) coefficient. First, key points are selected using principal component analysis and further extracted using the k-means clustering algorithm. Second, a grayscale image for each key point is generated as a feature descriptor using inverse distance weighted interpolation. Then, the presented method determines key point correspondences by improving the DIC coefficient. Finally, we apply the iterative closest point algorithm to fine registration. Experiment results show that the proposed method is accurate, time efficient, and it has a good antinoise performance.

COMPARING THE PERFORMANCE OF POINT CLOUD REGISTRATION METHODS FOR LANDSLIDE MONITORING USING MOBILE LASER SCANNING DATA

Abstract. The aim of the research is to evaluate the performance of the point cloud registration methods using mobile laser scanning data. The point cloud registration methods involved in this research are match bounding-box centres and iterative closest point (ICP). The research began with the two epoch’s mobile laser scanning survey using a Phoenix AL-3-32 system. At the same time, the stereo images of the study area were acquired using UAV Photogrammetric method. Both two epoch point cloud datasets were gone through the pre and post-processing stages to produce the cleaned and geo-referenced point clouds data. The data were then gone through the two registration methods and four Cloud-to-Cloud (C2C) distance methods. The 3D surface deviation results obtained from mobile laser scanning data was compared with the 3D surface deviation results from UAV data that undergoes the same registration and C2C distance computation methods. The study area involved in the research is an active landslide area that was located at Kulim Hi-Tech residential area in Kedah state, Malaysia. The study area exposed to the movement of the land which caused cracked to the buildings and drainages. The findings show that the ICP registration becomes the most suitable method to register point clouds dataset that was acquired using mobile laser scanning system. Among the four C2C distance computation methods that was involved in the testing, the least square plane method was the best method to calculate the distance between two sets of point clouds datasets which in turn gave the best results in the process of detecting the movement of the land in the study area.

Hierarchical registration of unordered TLS point clouds based on binary shape context descriptor

Automatic registration of unordered point clouds collected by the terrestrial laser scanner (TLS) is the prerequisite for many applications including 3D model reconstruction, cultural heritage management, forest structure assessment, landslide monitoring, and solar energy analysis. However, most of the existing point cloud registration methods still suffer from some limitations. On one hand, most of them are considerable time-consuming and high computational complexity due to the exhaustive pairwise search for recovering the underlying overlaps, which makes them infeasible for the registration of large-scale point clouds. On the other hand, most of them only leverage pairwise overlaps and rarely use the overlaps between multiple point clouds, resulting in difficulty dealing with point clouds with limited overlaps. To overcome these limitations, this paper presents a Hierarchical Merging based Multiview Registration (HMMR) algorithm to align unordered point clouds from various scenes. First, the multi-level descriptors (i.e., local descriptor: Binary Shape Context (BSC) and global descriptor: Vector of Locally Aggregated Descriptor (VLAD)) are calculated. Second, the point clouds overlapping (adjacent) graph is efficiently constructed by leveraging the similarity between their corresponding VLAD vectors. Finally, the proposed method hierarchically registers multiple point clouds by iteratively performing optimal registration point clouds calculation, BSC descriptor based pairwise registration and point cloud groups overlapping (adjacent) graph update, until all the point clouds are aligned into a common coordinate reference. Comprehensive experiments demonstrate that the proposed algorithm obtains good performance in terms of successful registration rate, rotation error, translation error, and runtime, and outperformed the state-of-the-art approaches.

MULTIPLE TLS POINT CLOUD REGISTRATION BASED ON POINT PROJECTION IMAGES

Abstract. In this paper, an efficient and robust registration method of multiple point clouds is proposed. In our research, we assume that point clouds are acquired by Terrestrial Laser Scanning (TLS) systems, and the scanned environments have a relatively flat base plane such as the ground or a floor. Our method is based on an existing pairwise registration method based on point projection images, which can quickly register the point clouds under the above assumptions. In the method, sliced point clouds are projected onto the base plane, and a binary image with feature points is created. The registration is done by using feature points of the images based on the sample consensus strategy. In this paper, first, we improve the efficiency of the pairwise registration method by introducing height and occlusion information to the image. Then, a validity check method of pairwise registration using space-classified images is proposed to avoid exhaustive pairwise registration in the multiple point cloud registration process. Finally, an efficient multiple point cloud registration algorithm based on progressive creation of a point cloud connectivity graph using iterative rough and precise pairwise registration and the validity check method is proposed. The effectiveness of our method is shown through its application to three datasets of outdoor environments.

Efficient Registration of Laser-Scanned Point Clouds of Bridges Using Linear Features

Efficient registration of point clouds from terrestrial laser scanners enables us to move from scanning to point cloud applications immediately. In this paper, a new efficient rough registration method of laser-scanned point clouds of bridges is proposed. Our method relies on straight-line edges as linear features, which often appear in many bridges. Efficient edge-line extraction and line-based registration methods are described in this paper. In our method, first, sampled regular point clouds based on the azimuth and elevation angles are created, and planar regions are extracted using the region growing on the regular point clouds. Then, straight lines of the edges of the planar regions are extracted as linear features. Next, vertical and horizontal line clusters are created according to the direction of the lines. To align the position and orientation of two point clouds, two corresponding nonparallel line pairs from line clusters are used. In the registration process, the RANSAC approach with a hash table of line pairs is used. In this process, the hash table is used for finding candidates of corresponding line pairs efficiently. Sampled points on the line pairs are used to align the line pairs, and occupied voxels and downsampled point clouds are used for efficient consensus calculation. The method is tested using three data sets of different types of bridges: a small steel bridge, a middle-size concrete bridge, and a high-pier concrete bridge. In our experiments, successful rates of our rough registration were 100%, and the processing time of rough registration for 19 point clouds was about 1 min.

基于激光强度分类的机载与地面激光雷达点云配准方法

基于激光强度分类的机载与地面激光雷达点云配准方法

An Automated Calibration Method of Ultrasonic Probe Based on Coherent Point Drift Algorithm

Ultrasound-based navigation, as a non-invasive and non-radiation image guiding system, is becoming a research focus in minimally invasive navigation surgery, and calibration between an ultrasonic probe and a 3D vision device is one of the key technologies for ultrasonic registration-based navigation. In this paper, a phantom model was designed as a benchmark for calibration. Both iterative closet point (ICP) and coherent point drift (CPD) algorithms are chosen as point cloud registration methods to implement calibration between ultrasonic scanned points and original phantom points to set up the relationship between the ultrasonic probe and the 3D vision device. Because of large topological difference between the ultrasound scanned points and the points from model, ICP algorithm cannot complete the registration, but the CPD algorithm could implement the registration automatically. The average errors of the center point position for each cavity were 1.50, 1.31, and 1.19 mm, respectively, and the average errors of the axis for each cavity were 0.85°, 0.61°, and 0.99°, respectively. Experiment results showed that the average error of calibration by this method satisfies acquirements of most orthopedic surgeries, and the fully automatic implementation of ultrasonic image processing and subsequent calculation is suitable for on-line calibration and verification in surgery.