Real Point Cloud Data Research Articles

Enhancing point cloud semantic segmentation in the data‐scarce domain of industrial plants through synthetic data

AbstractDigitizing existing structures is essential for applying digital methods in architecture, engineering, and construction. However, the adoption of data‐driven techniques for transforming point cloud data into useful digital models faces challenges, particularly in the industrial domain, where ground truth datasets for training are scarce. This paper investigates a solution leveraging synthetic data to train data‐driven models effectively. In the investigated industrial domain, the complex geometry of building elements often leads to occlusions, limiting the effectiveness of conventional sampling‐based synthetic data generation methods. Our approach proposes the automatic generation of realistic and semantically enriched ground truth data using surface‐based sampling methods and laser scan simulation on industry‐standard 3D models. In the presented experiments, we use a neural network for point cloud semantic segmentation to demonstrate that compared to sampling‐based alternatives, simulation‐based synthetic data significantly improve mean class intersection over union performance on real point cloud data, achieving up to 7% absolute increase.

A case study of quality control application with BIM-laser scanning collaboration in building construction process

Housing accounts for the largest share of the volume of the construction sector. Housing is the living space that people use on a permanent or temporary basis. Living space is also expected to have the desired quality. Quality requires construction in accordance with specified conditions and rules. In order to minimize possible errors in the design and construction process and to achieve the desired level of quality, continuous, scientific and instrumental controls are required. In recent years, with the rapid development of technology, methods have been developed that have achieved good results using integrated technology, which has begun to replace the quality controls performed by traditional methods. This new technology partnership called Building Information Modelling and Laser Scanning is described in this study. The aim of the study is to show that human error in building inspection can be minimized with the help of technology. In the study, a building was selected from a sample public housing project. Two dimensional projects of the building were converted into three dimensions using Building Information Modelling. In the current state of construction, point clouds were captured using laser scanning. The point clouds were converted to three dimensions. The construction defects were calculated by overlaying the BIM model with the real point cloud data. After checking, it was found that 2.2 % of the productions were defective productions. It was found that 97.8 % of the productions were within acceptable tolerance limits. The results obtained on a sample residential building demonstrate the importance of these new solutions for quality control and error-free production

A Real 3D Embodied Dataset for Robotic Active Visual Learning

Active interaction with environments is one of the striking characteristics of robotic active vision, which allows robots to move to facilitate visual tasks. Recently, several embodied AI platforms have been proposed as the synthetic environments to study robotic active vision, without needing to interact in real world. However, by using synthetic data, model trained on these platforms will suffer performance degradation when applied in reality. In this letter, a real 3D embodied dataset is proposed for robotic active visual learning. The proposed dataset consists of real point cloud data densely collected in 7 real-world indoor scenes. In our embodied dataset, researchers are able to simulate the movements and interactions of robots in indoor environments and obtain real visual data, which will not lead to performance degradation in reality. Furthermore, we proposed a 3D divergency policy that can guide robots to move and collect data to improve visual performance in novel environments. The proposed policy is designed following a simple fact: a good 3D detector should produce consistent 3D detection results for the same object from different viewpoints. Therefore, our policy encourages the robot to explore the area where the detector generates different 3D bounding boxes for the same object and helps the robot improve its visual performance in novel scenes.

Simulation of tree point cloud based on the ray-tracing algorithm and three-dimensional tree model

A terrestrial laser scanning (TLS) instrument can scan trees to get three-dimensional (3D) point clouds with detailed tree structures. However, some factors, such as surrounding environment, tree structure, the characteristics and scanning parameters of TLS instrument, can affect the scanning result. Simulation of tree point cloud is an efficient way to avoid and analyse the influence of the above factors. A simulation method was proposed to simulate tree point clouds by using the ray-tracing algorithm and 3D tree models. Thirteen tree models were tested with the proposed simulation method. Ten tree models are from the virtual model library, and other three tree models are constructed from real point cloud data. The accuracy of the simulated results was evaluated by using the absolute value of Hausdorff distances (AVOHD). Although the results showed that leaves and more branches will decrease simulation accuracy, the proposed method is effective and accurate on tree point cloud simulation.

An Innovative Detection Method of High-Speed Railway Track Slab Supporting Block Plane Based on Point Cloud Data from 3D Scanning Technology

The dimension detection of high-speed railway track slabs is one of the most important tasks before the track slabs delivery. Based on the characteristics of a 3D scanner which can acquire a large amount of measurement data continuously and rapidly in a short time, this paper uses the integration of 3D scanner and the intelligent robot to detect the CRTSIII (China Railway Track System) track slab supporting block plane, then the dense and accurate supporting block plane point cloud data is obtained, and the point cloud data is registered with the established model. An improved Random Sample Consensus (RANSAC) plane fitting algorithm is also proposed to extract the data of supporting block plane point cloud in this paper. The detection method is verified and the quality analysis of the detection results is assessed by a lot of real point cloud data obtained on site. The results show that the method can meet the quality control of CRTSIII finished track slab and the detection standard. Compared with the traditional detection methods, the detection method proposed in this paper can complete the detection of a track slab in 7 min, which greatly improves the detection efficiency, and has better reliability. The method has wide application prospects in the field of railway component detection.

PCBC: Quasiconformality of Point Cloud Mappings

The study of surface mappings or deformations plays an important role for various applications in computer visions and graphics. An accurate and effective method to measure and control geometric distortions of the mapping is therefore necessary. Quasiconformality, which captures the local geometric distortion of a surface mapping, is a useful tool for this purpose. In discrete setting, surfaces are often represented by point clouds. Although quasiconformal theories are well developed in the continuous setting, the concept of quasiconformality on point clouds is still lacking. In this paper, we propose a geometric quantity, called the Point Cloud Beltrami Coefficient (PCBC), to measure quasiconformality of a point cloud mapping. Its ability to measure the local geometric distortion of a mapping is theoretically and numerically validated. Moreover, we propose an algorithm to solve for a point cloud mapping with a prescribed PCBC. Numerical experiments have been carried out on both synthetic and real point cloud data, which demonstrate the efficacy of the proposed algorithm.

ROBUST CYLINDER FITTING IN THREE-DIMENSIONAL POINT CLOUD DATA

Abstract. This paper investigates the problems of cylinder fitting in laser scanning three-dimensional Point Cloud Data (PCD). Most existing methods require full cylinder data, do not study the presence of outliers, and are not statistically robust. But especially mobile laser scanning often has incomplete data, as street poles for example are only scanned from the road. Moreover, existence of outliers is common. Outliers may occur as random or systematic errors, and may be scattered and/or clustered. In this paper, we present a statistically robust cylinder fitting algorithm for PCD that combines Robust Principal Component Analysis (RPCA) with robust regression. Robust principal components as obtained by RPCA allow estimating cylinder directions more accurately, and an existing efficient circle fitting algorithm following robust regression principles, properly fit cylinder. We demonstrate the performance of the proposed method on artificial and real PCD. Results show that the proposed method provides more accurate and robust results: (i) in the presence of noise and high percentage of outliers, (ii) for incomplete as well as complete data, (iii) for small and large number of points, and (iv) for different sizes of radius. On 1000 simulated quarter cylinders of 1m radius with 10% outliers a PCA based method fit cylinders with a radius of on average 3.63 meter (m); the proposed method on the other hand fit cylinders of on average 1.02 m radius. The algorithm has potential in applications such as fitting cylindrical (e.g., light and traffic) poles, diameter at breast height estimation for trees, and building and bridge information modelling.

Geometric Priors for Gaussian Process Implicit Surfaces

This paper presents an extension of Gaussian process implicit surfaces (GPIS) by the introduction of geometric object priors. The proposed method enhances the probabilistic reconstruction of objects from three-dimensional (3-D) pointcloud data, providing a rigorous way of incorporating prior knowledge about objects expected in a scene. The key ideas, including the systematic use of surface normal information, are illustrated with one-dimensional and two-dimensional examples, and then applied to simulated and real pointcloud data for 3-D objects. The performance of our method is demonstrated in two different application scenarios, using complete and partial surface observations. Qualitative and quantitative analysis of the results reveals the superiority of the proposed approach over existing GPIS configurations that do not exploit prior knowledge.

An adaptive approach for the reconstruction and modeling of as-built 3D pipelines from point clouds

Abstract Automated extraction of 3D geometric shapes such as planes, spheres, cylinders, cones, and tori in laser-scanned point clouds is a challenging problem and a tedious process, especially when using cluttered data. This paper describes a modification of the existing Hough transform for the automatic detection of cylinder parameters in point clouds. Careful analysis reveals that the existing method still has excessive space and time complexity or yields imprecise outcomes. The approach described here modifies the orientation estimation with an area-based adaptive method that utilizes a small accumulator to detect significant peaks in the Hough space in the presence of single or multiple cylinders in the point cloud data. After orientation estimation, the position and radius are estimated using an orthonormal coordinate system with a circle fitting algorithm. These modifications are tested with extensive sets of real point cloud data, and experimental results show that the presented approach minimizes the space and time complexity. After detection, the relationship between cylinders is reconstructed to form a continuous axis network by tracking cylinder parameters obtained from earlier steps. Using the axis network of cylinders obtained from point clouds, models of entire pipelines that include straight pipes, elbow joints, and T-junctions are determinately defined, and output data is reconstructed in Smart Plant 3D (SP3D). The presented results show that the proposed approach indeed improves the computational complexity by reducing the space and time, and yields methods that can be employed in the automation of 3D pipeline model reconstruction.

CHANGE DETECTION OF MOBILE LIDAR DATA USING CLOUD COMPUTING

Change detection has long been a challenging problem although a lot of research has been conducted in different fields such as remote sensing and photogrammetry, computer vision, and robotics. In this paper, we blend voxel grid and Apache Spark together to propose an efficient method to address the problem in the context of big data. Voxel grid is a regular geometry representation consisting of the voxels with the same size, which fairly suites parallel computation. Apache Spark is a popular distributed parallel computing platform which allows fault tolerance and memory cache. These features can significantly enhance the performance of Apache Spark and results in an efficient and robust implementation. In our experiments, both synthetic and real point cloud data are employed to demonstrate the quality of our method.

CHANGE DETECTION OF MOBILE LIDAR DATA USING CLOUD COMPUTING

Abstract. Change detection has long been a challenging problem although a lot of research has been conducted in different fields such as remote sensing and photogrammetry, computer vision, and robotics. In this paper, we blend voxel grid and Apache Spark together to propose an efficient method to address the problem in the context of big data. Voxel grid is a regular geometry representation consisting of the voxels with the same size, which fairly suites parallel computation. Apache Spark is a popular distributed parallel computing platform which allows fault tolerance and memory cache. These features can significantly enhance the performance of Apache Spark and results in an efficient and robust implementation. In our experiments, both synthetic and real point cloud data are employed to demonstrate the quality of our method.

Using geometry tessellation and Markov chain Monte Carlo for segmentation of LiDAR point cloud data

This paper presents a segmentation algorithm of LiDAR point cloud data by which geometric and distributional features of objects are extracted. In this proposed algorithm, each object is considered to occupy a statistically homogeneous region and its acquired elevations are modeled as a normal distribution. To segment the LiDAR point cloud into homogeneous regions, a Voronoi tessellation is first used to partition its domain into polygons. The number of polygons is given in practice. Each of the polygons is assigned a random label variable to indicate the region to which it belongs. By Bayesian inference, the joint probability of labels and distribution parameters conditional on the given dataset can be obtained up to a normalizing constant. A Markov chain Monte Carlo scheme is designed to simulate from the posterior and to estimate the model parameters. Finally, the optimal segmentation is obtained under maximum <italic<a posteriori</italic< estimation. Experiments on real point cloud data show that normal distribution parameters for each region quickly converge to their stable values, and the optimal segmentation results can be obtained within 20,000 iterations for all datasets. Experiments on simulated point cloud data demonstrate that the proposed algorithm can accurately estimate the model parameters.

Rock Surface Classification in a Mine Drift Using Multiscale Geometric Features

Scale-dependent statistical depictions of surface morphology offer the potential to parameterize complex geometrical scaling relationships with greater detail than traditional fractal measures. Using multiscale operators, it is possible to identify points belonging to rough discontinuous surfaces in noisy point clouds solely on the basis of their local geometry. Many strategies for point cloud feature classification have been developed since the proliferation of laser scanning systems. Most of the techniques which are applicable to natural scenes employ external data sources such as hyperspectral imagery, return pulse intensity, and waveform data. In this letter, multiscale geometric parameters are used to identify individual point observations corresponding to rock surfaces in point clouds acquired by terrestrial laser scanning in scenes with man-made clutter and scanning artifacts. Three multiscale operators, namely, the approximate shape and density of a defined neighborhood and the distance of its mean point from its geometric center, are fused into a single feature vector. The procedure is demonstrated using real point cloud data acquired in a mine drift, with the goal of identifying points belonging to the rock face obscured by an overlying wire support mesh. Using the extra-trees classifier, extraneous returns caused by the mesh were excluded from the point cloud with a 97% success rate, while 87% of the desired surface points were retained.