Global Registration of 3D LiDAR Point Clouds Based on Scene Features: Application to Structured Environments

Abstract

Acquiring 3D data with LiDAR systems involves scanning multiple scenes from different points of view. In actual systems, the ICP algorithm (Iterative Closest Point) is commonly used to register the acquired point clouds together to form a unique one. However, this method faces local minima issues and often needs a coarse initial alignment to converge to the optimum. This paper develops a new method for registration adapted to indoor environments and based on structure priors of such scenes. Our method works without odometric data or physical targets. The rotation and translation of the rigid transformation are computed separately, using, respectively, the Gaussian image of the point clouds and a correlation of histograms. To evaluate our algorithm on challenging registration cases, two datasets were acquired and are available for comparison with other methods online. The evaluation of our algorithm on four datasets against six existing methods shows that the proposed method is more robust against sampling and scene complexity. Moreover, the time performances enable a real-time implementation.Data Set: https://liris.cnrs.fr/3d-registration/ Data Set License: ODC Attribute License

Image-Based Method for the Pairwise Registration of Mobile Laser Scanning Point Clouds

“…The use of point cloud data retrieved from MLS techniques requires the integration of all the 3D scans collected at different times from different observation points in a common reference system, a process known as global registration (Sanchez et al, 2017). However, prior to that it is necessary to perform relative alignment of the overlapping point clouds to facilitate the construction of a single 3D point cloud model.…”

“…Also, ICP-based methods perform poorly when points in one scan do not have correspondences in the other (Pomerleau et al, 2013), which is very common in MLS data (Figure 1a and 1b). Furthermore, this algorithm delivers incorrect results if the initial position of the point clouds is not close to the required matching position, or simply put the offset is large ( (Shetty, 2017), (Sanchez et al, 2017)). This can be the case in MLS data when streets separate dense blocks of high structures or when tall trees are present, as then the GNSS reception is highly restricted.…”

“…The algorithm establishes iteratively correspondences between the points of two point cloud sets and computes the spatial distances between them. It terminates when the sum of the spatial distances between the correspondences is minimum (Sanchez et al, 2017).…”

Image-Based Vehicle Tracking From Roadside Lidar Data

“…Byun et al, 2017). The algorithm iteratively establishes correspondences between the points of two point cloud datasets, computes the spatial distances between them and terminates when the sum of the spatial distances between the correspondences is at a minimum (Sanchez et al, 2017). However, this approach is computationally expensive since it requires an extensive search of point correspondences between the point clouds (Godin et al, 1994).…”

“…However, this approach is computationally expensive since it requires an extensive search of point correspondences between the point clouds (Godin et al, 1994). Moreover, this method faces local minima issues and often needs a coarse initial alignment to converge to the optimum solution (Sanchez et al, 2017). To avoid such problems in 3D point cloud registration, 2D image-based point cloud registration approaches have been proposed in some studies (e.g.…”