Real-time 3D point cloud registration

As a fundamental process of three-dimensional lidar point cloud data (3D LPCD) processing, numerous registration methods are time consuming and easily fall into local optimum. A 3D LPCD registration method based on the iterative closest point (ICP) algorithm, which is improved by the Gaussian mixture model (GMM) considering corner features, is proposed in this article to address these limitations. The GMM method is used for coarse registration, and the input original 3D LPCD is replaced by corner features extracted by the improved 3D Harris algorithm to improve the efficiency of coarse registration. In addition, a satisfactory initial position between the reference and the moving 3D LPCD is prepared for ICP fine registration by coarse registration; thus, the accuracy of fine registration can be improved. The registration accuracy and efficiency of the new method is proved to be higher than those of four common ICP-based registration methods (3DSC-RANSACICP, 3DSC-SAC-IAICP, FPFH-RANSACICP, and FPFH-SAC-IAICP), and GMM registration methods, and the local optimum problem is effectively addressed.

Abstract. Laser scanning samples the surface geometry of objects efficiently and records versatile information as point clouds. However, often more scans are required to fully cover a scene. Therefore, a registration step is required that transforms the different scans into a common coordinate system. The registration of point clouds is usually conducted in two steps, i.e. coarse registration followed by fine registration. In this study an automatic marker-free coarse registration method for pair-wise scans is presented. First the two input point clouds are re-sampled as voxels and dimensionality features of the voxels are determined by principal component analysis (PCA). Then voxel cells with the same dimensionality are clustered. Next, the Extended Gaussian Image (EGI) descriptor of those voxel clusters are constructed using significant eigenvectors of each voxel in the cluster. Correspondences between clusters in source and target data are obtained according to the similarity between their EGI descriptors. The random sampling consensus (RANSAC) algorithm is employed to remove outlying correspondences until a coarse alignment is obtained. If necessary, a fine registration is performed in a final step. This new method is illustrated on scan data sampling two indoor scenarios. The results of the tests are evaluated by computing the point to point distance between the two input point clouds. The presented two tests resulted in mean distances of 7.6 mm and 9.5 mm respectively, which are adequate for fine registration.

In response to the issues of slow convergence and the tendency to fall into local optima in traditional iterative closest point (ICP) point cloud registration algorithms, this study presents a fast registration algorithm for laser point clouds based on 3D scale-invariant feature transform (3D-SIFT) feature extraction. First, feature points are preliminarily extracted using a normal vector threshold; then, more high-quality feature points are extracted using the 3D-SIFT algorithm, effectively reducing the number of point cloud registrations. Based on the extracted feature points, a coarse registration of the point cloud is performed using the fast point feature histogram (FPFH) descriptor combined with the sample consensus initial alignment (SAC-IA) algorithm, followed by fine registration using the point-to-plane ICP algorithm with a symmetric target function. The experimental results show that this algorithm significantly improved the registration efficiency. Compared with the traditional SAC-IA+ICP algorithm, the registration accuracy of this algorithm increased by 29.55% in experiments on a public dataset, and the registration time was reduced by 81.01%. In experiments on actual collected data, the registration accuracy increased by 41.72%, and the registration time was reduced by 67.65%. The algorithm presented in this paper maintains a high registration accuracy while greatly reducing the registration speed.

Incremental point clouds registration is studied in this paper. A rapid method for point clouds registration based on reference points is proposed, which consists of the coarse registration and fine registration. Firstly, a set of reference points is applied as an assistant utility to measure the object. The transformation parameters are estimated by using the reference points only for coarse registration, and then dense point clouds data will be transformed to the same coordinate system. Secondly, taking the coarse registration results as the initial value, the improved Interactive Closest Point (ICP) algorithm is used in fine registration the original corresponding points are established rapidly by using the k-d tree searching algorithm. Finally, Preview Model Parameters Evaluation Random Sample Consensus (PERANSAC) algorithm is utilized to remove outliers. The experimental result shows that this method in finding original corresponding points can greatly improve the computation efficiency and also improve the registration accuracy.

Surface registration is widely used in image-guided neurosurgery to achieve spatial registration between the patient space and the image space. Coarse registration, followed by fine registration, is an important premise to ensure the robustness and efficiency of surface registration. In this paper, a coarse registration algorithm based on the principal axes is proposed to achieve this goal. The extraction of the principal axes relies on the approximated surface with an adaptive Gaussian kernel, the width of which is consistent with neighborhood relation so that it is applicable for various scanning data. Determining the corresponding centers of translation is another problem for aligning different scanning data, which is solved through heuristics. Six pairs of points on two surfaces with the farthest projections on the principal axes were regarded as the candidates of translation centers, and then through tentative alignments of local regions around them, a pair of candidates with the minimum registration error was selected as the optimal translation centers. Automatic registration of two scans of a head phantom is presented in this paper. Experimental results confirmed the robustness of the algorithm and its feasibility in clinical applications.

Evaluation of point cloud registration using Monte Carlo method

Nowadays precast concrete elements are widely used in buildings, bridges, and other civil infrastructure facilities because precast elements allow rapid construction and high precision quality control. However, the quality inspection of precast concrete elements primarily relies on manual inspection, which is time consuming and error-prone. Recently, terrestrial laser scanner (TLS) has been used to improve quality inspection. The authors’ group has previously developed an automated dimension estimation technique for precast concrete elements, but its applicability is limited only to precast elements with rectangular shapes. This study advances our previous work so that the dimensions of precast elements with irregular shapes can also be automatically estimated. First, a density-based clustering algorithm is adopted to extract target objects from 3D point cloud data acquired by a TLS. Then, coarse registration is conducted to match each object extracted from the 3D point cloud with the asdesign objects in building information models (BIM) one by one. Thirdly, all point cloud data points are registered onto different surfaces of the as-designed objects in BIM through fine registration. Lastly, the as-built dimensions of the precast concrete element are extracted and compared with the as-design ones in BIM. The effectiveness and accuracy of the developed technique are examined using point cloud data obtained from a laboratory-scale precast concrete bridge deck panel.

Multisensor systems with optical 3D sensors are frequently employed to capture complete surface information by measuring workpieces from different views. During coarse and fine registration the resulting datasets are afterward transformed into one common coordinate system. Automatic fine registration methods are well established in dimensional metrology, whereas there is a deficit in automatic coarse registration methods. The advantage of a fully automatic registration procedure is twofold: it enables a fast and contact-free alignment and further a flexible application to datasets of any kind of optical 3D sensor. In this paper, an algorithm adapted for a robust automatic coarse registration is presented. The method was originally developed for the field of object reconstruction or localization. It is based on a segmentation of planes in the datasets to calculate the transformation parameters. The rotation is defined by the normals of three corresponding segmented planes of two overlapping datasets, while the translation is calculated via the intersection point of the segmented planes. First results have shown that the translation is strongly shape dependent: 3D data of objects with non-orthogonal planar flanks cannot be registered with the current method. In the novel supplement for the algorithm, the translation is additionally calculated via the distance between centroids of corresponding segmented planes, which results in more than one option for the transformation. A newly introduced measure considering the distance between the datasets after coarse registration evaluates the best possible transformation. Results of the robust automatic registration method are presented on the example of datasets taken from a cutting tool with a fringe-projection system and a focus-variation system. The successful application in dimensional metrology is proven with evaluations of shape parameters based on the registered datasets of a calibrated workpiece.

In the recent era, a lot of interest is attracted by the autonomous vehicles which can sense surroundings and navigate without human intervention. Object detection and recognition form a major part of autonomous driving systems. Lidar sensors can be used to capture point clouds of driving environment. Detecting multiple 3D objects in point clouds in real time and defining their boundaries with the help of 3D bounding boxes are critical in motion planning by self driving systems. This paper proposes a LiDAR-based 3D object detection system that operates in real-time, with emphasis on autonomous driving scenarios. A state-of-the-art 2D standard object detector for RGB images, YOLOv4, is used as the base for object detection. The multi-class 3D bounding boxes are generated using a complex regression approach. An Euler-Region-Proposal Network (E-RPN) is used to predict the pose of the object. The proposed model receives point cloud data as input and outputs 3D bounding boxes with classes in real-time. The experiments done on the KITTI benchmark dataset proves that the proposed system outperforms existing methods in terms of accuracy and performance.

Accurate and efficient registration of unmanned aerial vehicle light detection and ranging (UAV‐lidar) and terrestrial lidar (T‐lidar) data is crucial for forest structure parameter extraction. This study proposes a novel method based on a starburst pattern for the automatic registration of UAV‐lidar and T‐lidar data in forest scenes. It employs density‐based spatial clustering of applications with noise (DBSCAN) for individual tree identification, constructs starburst patterns separately from both lidar sources, and utilises polar coordinate rotation and matching to achieve coarse registration. Fine registration is achieved using the iterative closest point (ICP) algorithm. Experimental results demonstrate that the starburst‐pattern‐based method achieves the desired registration accuracy (average coarse registration error of 0.157 m). Further optimisation with ICP yields slight improvements with an average fine registration error of 0.149 m. Remarkably, the proposed method is insensitive to the individual tree detection number when exceeding 10, and the tree position error has minimal impact on registration accuracy. Furthermore, our proposed method outperforms two existing methods in T‐lidar and UAV‐lidar registration over forest environments.

Point cloud registration is a central theme in computer vision, with alignment algorithms continuously improving for greater robustness. Commonly used methods evaluate Euclidean distances between point clouds and minimize an objective function, such as Root Mean Square Error (RMSE). However, these approaches are most effective when the point clouds are well-prealigned and issues such as differences in density, noise, holes, and limited overlap can compromise the results. Traditional methods, such as Iterative Closest Point (ICP), require choosing one point cloud as fixed, since Euclidean distances lack commutativity. When only one point cloud has issues, adjustments can be made, but in real scenarios, both point clouds may be affected, often necessitating preprocessing. The authors introduce a novel differential entropy-based metric, designed to serve as the objective function within an optimization framework for fine rigid pairwise 3D point cloud registration, denoted as Iterative Differential Entropy Minimization (IDEM). This metric does not depend on the choice of a fixed point cloud and, during transformations, reveals a clear minimum corresponding to the best alignment. Multiple case studies are conducted, and the results are compared with those obtained using RMSE, Chamfer distance, and Hausdorff distance. The proposed metric proves effective even with density differences, noise, holes, and partial overlap, where RMSE does not always yield optimal alignment.

In this paper, we present a simple, efficient, and robust algorithm for 2D coarse registration of two point clouds. In the proposed algorithm, the locations of some distinct objects are detected from the point cloud data, and a rotation- and translation-invariant feature descriptor vector is computed for each of the detected objects based on the relative locations of the neighboring objects. Subsequently, the feature descriptors obtained for the different point clouds are compared against one another by using the Euclidean distance in the feature space as the similarity criterion. By using the nearest neighbor distance ratio, the most promising matching object pairs are found and further used to fit the optimal Euclidean transformation between the two point clouds. Importantly, the time complexity of the proposed algorithm scales quadratically in the number of objects detected from the point clouds. We demonstrate the proposed algorithm in the context of forest inventory by performing coarse registration between terrestrial and airborne point clouds. To this end, we use trees as the objects and perform the coarse registration by using no other information than the locations of the detected trees. We evaluate the performance of the algorithm using both simulations and three test sites located in a boreal forest. We show that the algorithm is fast and performs well for a large range of stem densities and for test sites with up to 10 ​000 trees. Additionally, we show that the algorithm works reliably even in the case of moderate errors in the tree locations, commission and omission errors in the tree detection, and partial overlap of the data sets. We also demonstrate that additional tree attributes can be incorporated into the proposed feature descriptor to improve the robustness of the registration algorithm provided that reliable information of these additional tree attributes is available. Furthermore, we show that the registration accuracy between the terrestrial and airborne point clouds can be significantly improved if stem positions estimated from the terrestrial data are matched to stem positions obtained from the airborne data instead of matching them to tree top positions estimated from the airborne data. Even though the 2D coarse registration algorithm is demonstrated in the context of forestry, the algorithm is not restricted to forest data and it may potentially be utilized in other applications, in which efficient 2D point set registration is needed.

In surgical navigation-assisted pedicle screw fixation, cross-source pre- and intra-operative point clouds registration faces challenges like significant initial pose differences and low overlapping ratio. Classical algorithms based on feature descriptor have high computational complexity and are less robust to noise, leading to a decrease in accuracy and navigation performance. To address these problems, this paper proposes a coarse-to-fine registration framework. In the coarse registration stage, a Point Matching algorithm based on Curvature Feature Learning (CFL-PM) is proposed. Through CFL-PM and Farthest Point Sampling (FPS), the coarse registration of overlapping regions between the two point clouds is achieved. In the fine registration stage, the Iterative Closest Point (ICP) is used for further optimization. The proposed method effectively addresses the challenges of noise, initial pose and low overlapping ratio. In noise-free point cloud registration experiments, the average rotation and translation errors reached 0.34° and 0.27 mm. Under noisy conditions, the average rotation error of the coarse registration is 7.28°, and the average translation error is 9.08 mm. Experiments on pre- and intra-operative point cloud datasets demonstrate the proposed algorithm outperforms the compared algorithms in registration accuracy, speed, and robustness. Therefore, the proposed method can achieve the precise alignment of the surgical navigation-assisted pedicle screw fixation.

An automatic 3D point cloud registration method based on regional curvature maps

Three-dimensional (3D) registration is the process aligning the range data sets form different views in a common coordinate system. In order to generate a complete 3D model, we need to refine the data sets after coarse registration. One of the most popular refinery techniques is the iterative closest point (ICP) algorithm, which starts with pre-estimated overlapping regions. This paper presents an improved ICP algorithm that can automatically register multiple 3D data sets from unknown viewpoints. The sensor projection that represents the mapping of the 3D data into its associated range image and a cross projection are used to determine the overlapping region of two range data sets. By combining ICP algorithm with the sensor projection, we can make an automatic registration of multiple 3D sets without pre-procedures that are prone to errors and any mechanical positioning device or manual assistance. The experimental results demonstrated that the proposed method can achieve more precise 3D registration of a couple of 3D data sets than previous methods.