Geometry Preserving Sampling Method Based on Spectral Decomposition for Large-Scale Environments

Abstract

In the context of 3D mapping, larger and larger point clouds are acquired with lidar sensors. Although pleasing to the eye, dense maps are not necessarily tailored for practical applications. For instance, in a surface inspection scenario, keeping geometric information such as the edges of objects is essential to detect cracks, whereas very dense areas of very little information such as the ground could hinder the main goal of the application. Several strategies exist to address this problem by reducing the number of points. However, they tend to underperform with non-uniform density, large sensor noise, spurious measurements, and large-scale point clouds, which is the case in mobile robotics. This paper presents a novel sampling algorithm based on spectral decomposition analysis to derive local density measures for each geometric primitive. The proposed method, called Spectral Decomposition Filter (SpDF), identifies and preserves geometric information along the topology of point clouds and is able to scale to large environments with a non-uniform density. Finally, qualitative and quantitative experiments verify the feasibility of our method and present a large-scale evaluation of SpDF with other seven point cloud sampling algorithms, in the context of the 3D registration problem using the Iterative Closest Point (ICP) algorithm on real-world datasets. Results show that a compression ratio up to 97 % can be achieved when accepting a registration error within the range accuracy of the sensor, here for large scale environments of less than 2 cm.

“Canopy fingerprints” for characterizing three-dimensional point cloud data of soybean canopies

“…Due to the variable distances between the scanner and the plots, the point cloud density for a single plot captured from four different corners/perspectives varied ( Figure 3B ). This disparity can cause problems in mesh generation and skeletonization ( Labussière et al, 2020 ; Xia et al., 2020 ). To reduce the disparity in point density, we voxelized the point clouds.…”

Optimising Faster R-CNN Training to Enable Video Camera Compression for Assisted and Automated Driving Systems

“…in excess of 40 Gb/s) for higher levels of autonomy (levels 3-5), current wired vehicle data networks are inadequate to reliably transmit the required data amount [3]- [5]. With the increased demand of automotive cameras providing high resolution (8)(9)(10)(11)(12) and the required high dynamic range (HDR) to cope with the luminosity variations when driving (e.g. bright sun in front of the sensor when travelling in a dark tunnel), cameras can significantly contribute to the amount of generated data by the sensor suite; moreover multiple cameras are required to provide 360°coverage of a vehicle's surroundings.…”

“…It is worth noting that a considerable amount of research is developing also in the area of LiDAR data reduction, but it is outside the scope of this paper. A good coverage of some of the challenges associated with pointcloud reduction can be found in [8].…”

Self-supervised Learning of LiDAR Odometry for Robotic Applications

“…More complex features [4][5][6] can be used to make the point selection process invariant to sensor orientation and robot pose, however highcomputational cost makes them unsuitable for real-time robot operation. Furthermore, although using all available scan data may not be necessary, yet it has been shown that utilizing more scan data up to a certain extent can improve the quality of the scan-to-scan alignment process [7].…”