China
Africa
Explore chapters and articles related to this topic
Point cloud registration for bridge defect tracking in as-built models
Published in Joan-Ramon Casas, Dan M. Frangopol, Jose Turmo, Bridge Safety, Maintenance, Management, Life-Cycle, Resilience and Sustainability, 2022
J. Bush, J. Ninić, G. Thermou, J. Bennetts, S. Denton, L. Tachtsi, P. Hill
Point clouds may be obtained from a number of 2D images using a photogrammetry-based reconstruction technique, or through use of 3D scanning sensors such as laser or LiDAR which capture a depth map on site. Given two point clouds of the same physical object (such as the same bridge defect recorded during two inspections several years apart), it is not a trivial task to compare them. Even if the raw data was captured using the same equipment and on the same day, no two sets of raw data are identical (slight differences in camera pose, differences in the extent of physical object captured, random point sampling, etc.). Further variability from use of different sensors and reconstruction algorithms gives rise to the challenge of having to adjust the scale and the orientation of these point clouds in order to align them.
Real-time 3D reconstruction using SLAM for building construction
Published in Konstantinos Papadikis, Chee S. Chin, Isaac Galobardes, Guobin Gong, Fangyu Guo, Sustainable Buildings and Structures: Building a Sustainable Tomorrow, 2019
3D reconstruction is a process of capturing the geometry and appearance of an object. Point cloud models, mesh models and geometric models are three main models to represent reconstructed models, where the point cloud models are the basis. Point cloud is a set of data points including XYZ coordinates data and color data, which form the external surface of an object. 3D reconstruction is widely used in construction site, for example, 3D reconstruction of as-built model in construction site can be applied in construction progress tracking, geometry quality inspection, construction safety management etc. (Wang et al. 2019). There are two main 3D reconstruction methods to obtain 3D point cloud models, which are the image-based ranging algorithms method and the laser scanner-based method. Image-based ranging algorithms processes 2D images to get 3D point cloud model, while 3D laser scanner directly generates point cloud model by depth information based on round trip time of the laser beam/ray. Both two methods can only generate 3D point clouds off-line. Off-line generating point cloud models causes delay of detecting errors in the collecting data step. For example, engineers generate a model and find that there are some missing parts. They have to recollect the data of the missing parts. Therefore, it is better to generate a point cloud model during the data collecting step. In the present paper, we propose a real-time 3D reconstruction system by using simultaneous localization and mapping (SLAM) and evaluate this system with the image-based algorithm SfM and laser scanner.
Algorithmic approaches to BIM modelling from reality
Published in Yusuf Arayici, John Counsell, Lamine Mahdjoubi, Gehan Nagy, Soheir Hawas, Khaled Dewidar, Heritage Building Information Modelling, 2017
Ebenhaeser Joubert, Yusuf Arayici
Laser scanning, photogrammetry and LIDAR are popular methods to produce point clouds. Understanding the basic workings of the methods utilised to produce 3D data is important. This allows the reader to put the new approach towards data processing within the context of current physical technology. Different methods of data collection offer a wide range of strengths and weaknesses. Laser scanning, for instance, is relatively fast and accurate in comparison to photogrammetry but is affected by shadowing and leaves gaps in the data consequently. Using microwave scanning affords much freedom in the wavelength propagated.
Automated variance modeling for three-dimensional point cloud data via Bayesian neural networks
Published in IISE Transactions, 2023
Zhaohui Geng, Arman Sabbaghi, Bopaya Bidanda
The point cloud is a data type that describes the shape of a design or an object via a large set of points described by the three Cartesian coordinates in the space. It is a very common and flexible data structure in CAD design and modern computer graphical applications (Wu et al., 2008; Balaban et al., 2012). For example, the most widely-used file format for freeform objects in AM is Standard Tessellation Language (STL), which is described by a set of vertices in the point cloud and the edges connecting them. In addition, point cloud data constitute the de facto data type for geometric metrology. Conventional metrology instruments, such as laser scanners and Coordinate Measuring Machines (CMMs), capture geometric information of parts by measuring the coordinates of the surface points and coding them into point cloud data. More importantly, accurate point cloud collection, also known as digitization, is invariably considered the first and foremost step for every RE or metrology project.
Adaptively unsupervised seepage detection in tunnels from 3D point clouds
Published in Structure and Infrastructure Engineering, 2022
Kunyu Wang, Xianguo Wu, Heng Li, Fan Wang, Limao Zhang, Hongyu Chen
Point cloud segmentation aims to obtain the interest region, and there are two main kinds of segmentation algorithms, including feature-based and model-based algorithms (Ma & Liu, 2018; Matsuura, Hayano, Itakura, & Suzuki, 2019). As a commonly used point cloud segmentation model, the deep convolutional neural network has achieved good results in many fields. For instance, Xie et al. proposed a lightweight CNN structure for projection-based LiDAR point cloud semantic segmentation for autonomous driving (Xie, Bai, & Huang, 2021). Te et al. introduced a regularized graph convolutional neural network that directly consumes point clouds which is robust to both noise and point cloud density (Te, Hu, Zheng, & Guo, 2018). Zhang et al. implemented point cloud classification and segmentation by a hybrid feature convolutional neural network model (X. Zhang, Fu, Zhao, & Xu, 2020). Wang et al. proposed a neural network module dubbed EdgeConv suitable for CNN-based high-level tasks on point clouds, including classification and segmentation (Wang et al., 2019). These methods do achieve good segmentation results but require more prerequisites, including good and large datasets, long training time, and complex hyperparameter tuning. In tunnel seepage identification, these prerequisites are tough to achieve. Therefore a method is needed to be able to achieve seepage identification without preparation, which can buy more time for disease management on one hand and provide data sets for more accurate supervised learning efficiently on the other hand.
Cylinder-based simultaneous registration and model fitting of laser-scanned point clouds for accurate as-built modeling of piping system
Published in Computer-Aided Design and Applications, 2018
Ryota Moritani, Satoshi Kanai, Hiroaki Date, Masahiro Watanabe, Takahiro Nakano, Yuta Yamauchi
Point cloud registration is an indispensable processing used for aligning multiple partial scans captured from different scanner positions, mapping them to consistent coordinates. Two types of registration method are used: marker-based and marker-less. Marker-based methods attach a set of artificial fiducial markers to the object surface, then identify an appropriate alignment among them. These methods afford high accuracy and robustness, and are widely used in laser-scanning applications. As the fiducial marker, small planar plates with black and white patterns or spheres [28] are generally used. However, marker installation is a dangerous and time-consuming process, especially when working high within a piping system. The accuracy of registration depends on the number of markers installed by the operator, and their placement.