Registration of Multi-Sensor Bathymetric Point Clouds in Rural Areas Using Point-to-Grid Distances

Richard Boerner,Frank Steinbacher,Ludwig Hoegner,Uwe Stilla,Ramona Baran,Yusheng Xu

doi:10.3390/ijgi8040178

Richard Boerner, Frank Steinbacher + Show 4 more

Open Access

https://doi.org/10.3390/ijgi8040178

Copy DOI

Abstract

This article proposes a method for registration of two different point clouds with different point densities and noise recorded by airborne sensors in rural areas. In particular, multi-sensor point clouds with different point densities are considered. The proposed method is marker-less and uses segmented ground areas for registration.Therefore, the proposed approach offers the possibility to fuse point clouds of different sensors in rural areas within an accuracy of fine registration. In general, such registration is solved with extensive use of control points. The source point cloud is used to calculate a DEM of the ground which is further used to calculate point to raster distances of all points of the target point cloud. Furthermore, each cell of the raster DEM gets a height variance, further addressed as reconstruction accuracy, by calculating the grid. An outlier removal based on a dynamic threshold of distances is used to gain more robustness against noise and small geometry variations. The transformation parameters are calculated with an iterative least-squares optimization of the distances weighted with respect to the reconstruction accuracies of the grid. Evaluations consider two flight campaigns of the Mangfall area inBavaria, Germany, taken with different airborne LiDAR sensors with different point density. The accuracy of the proposed approach is evaluated on the whole flight strip of approximately eight square kilometers as well as on selected scenes in a closer look. For all scenes, it obtained an accuracy of rotation parameters below one tenth degrees and accuracy of translation parameters below the point spacing and chosen cell size of the raster. Furthermore, the possibility of registration of airborne LiDAR and photogrammetric point clouds from UAV taken images is shown with a similar result. The evaluation also shows the robustness of the approach in scenes where a classical iterative closest point (ICP) fails.

Highlights

The task of registration is to transfer two data sets of the same scene captured on different times or with different sensors into the same coordinate frame
While the whole flight strip shows a mostly rural area, the proposed method shows that no human-made structure is needed for registration
In comparison to the classical iterative closest point (ICP) [1], it is demonstrated that the proposed method can solve the registration where the ICP gets into trouble

Summary

Introduction

The task of registration is to transfer two data sets of the same scene captured on different times or with different sensors into the same coordinate frame. In the case of 3D data linear structures, planar structures, or point correspondences are used as geometric features. These features have the disadvantage that using point correspondences as used in the iterative closest point (ICP) [1] algorithm is limited to point clouds with almost the same point distribution and density. Using linear or planar structures derived from the point clouds overcome the limitation of different point densities, but these are mainly limited to scenes with human-made structures. In rural areas, these solutions are not applicable in most cases

Methods

Results

Discussion

Conclusion