Target-Free Automatic Registration of Point Clouds

Abstract

-The aim of this paper is to introduce a novel method that automatically registers colored 3D point cloud sets without using targets or any other manual alignment processes. For fully automated point cloud registration without targets or landmarks, our approach utilizes feature detection algorithms used in computer vision. A digital camera and a laser scanner is utilized and the sensor data is merged based on a kinematic solution. The proposed approach is to detect and extract common features not directly from a 3D point cloud but from digital images corresponding to the point clouds. The initial alignment is achieved by matching common SURF features from corresponding digital images. Further alignment is obtained using plane segmentation and matching from the 3D point clouds. The test outcomes show promising results in terms of registration accuracy and processing time.

Framework for digital twin creation in off-road environments from LiDAR scans

“…On the other hand, the robotic SLAM-driven method is advantageous for registering large volumes of scan data since it is fully autonomous with no manual input required. 42,43 In practice, we would recommend that TLS scanning be used in scenarios where high fidelity reconstructions are needed and robotic scanning be used where the region of interest is too large and a high level of automation is needed.…”

As-is Geometric Data Collection and 3D Visualization through the Collaboration between UAV and UGV

“…One method for producing a 3D sitemap is photogrammetry. Photographs provide useful information about the construction progress that can be automatically processed and converted to 3D point clouds using Structure from Motion [7]- [9]. Due to stability and payload issues, UAVs typically use a photo camera to capture scenes and build a 3D point cloud with photogrammetry.…”

“…However, multiple scans from different locations must be taken and registered because of limited data capture coverage and occlusions. To register multiple scans in one coordinate system, many physical targets or markers are typically pre-installed in the overlapping scan area, which requires substantial cost, labor, and time [9]. A well-designed scanning plan minimizes data collection time while capturing all required geometric information.…”

Robotic sensing and object recognition from thermal-mapped point clouds

“…There are two kinds of matrices for camera calibration; the camera calibration process finds both the internal and external parametric matrices for the camera, which affect the image processing. The internal parametric matrix consists of the intrinsic parameters, the focal length, image sensor format, and principal point, while the extrinsic parameters can be obtained through a geometric relationship based on the mounting configuration such as the height and direction of the camera (Kim et al 2016). Using these intrinsic and extrinsic parameters, the 3D laser-scanned point cloud data X, Y, Z in the body-fixed coordinate system which has origin at the center of the laser scanner robot body can be transformed into 3D camera coordinates (u, v, w) which has origin at the focus of a camera, in Eq.…”