FOSS4G 2022 academic track

In the recent years, point cloud technologies, such as Unmanned Aerial Vehicles (UAV), Terrestrial Laser Scanners (TLS), Aerial Laser Scanners (ALS), let alone Mobile Mapping Systems (MMS) have come into the focus of attention and have been a subject of considerable public concern in mapping.  Thanks to these new techniques, experts can survey large areas with sufficient and homogenous accuracy, with high resolution.

It comes from this that there are several areas where the point clouds can be used. One of the applications is updating land registry maps. Many countries all over the world face the issue that a significant part of their large-scale land registry maps are based on old analogue maps from the late 19th or the early 20th century. One of these countries is Hungary, where more than eighty percent of digital cadastre maps were digitised using analogous maps in a scale range of 1:1000 – 1:4000, not to mention the maps with fathom as base unit and with the scale of 1:2880. It is quite common to have a few meters offset in the features depicted in the land registry maps, which yields a wide variety of problems in applying maps, such as in public utility registration and engineering practice. The final solution to the problem would be to carry out new surveys for the critical areas, but that has been often postponed due to the lack of time and excessive costs.

Thanks to the new technologies updating the old and not relevant maps are feasible and there are several examples, where point clouds were used to update old land registry maps with manual processing. As it has been investigated by many researchers, an optimal solution is to generate point clouds from the combination of nadir and oblique images taken by UAVs, typically having 1-3 cm Ground Sample Distance (GSD). Our aim is to find the building footprints with not more than 10 cm accuracy from the point clouds. Oblique images play an important role in having sufficient number of points on the walls of the buildings in the point cloud, so we can find not only the outline of the roofs but the walls of the buildings, too.

There is another crucial factor that needs to be considered when processing point clouds, namely that of automation. It is beyond doubt that automation definitely improves the efficiency of the whole procedure. There is already a wide range of open-source software available, such as OpenDroneMap (ODM)/WebODM, CloudCompare, QGIS, not to mention many open-source libraries, like Open3D, PDAL, Point Cloud Library (PCL), SciPy and Scikit-learn to support automatic data processing of point clouds. During our research, the different combination of these libraries was investigated paying attention to be accessible and freely developable for everyone. Therefore, the source code (mostly written in Python) of our programs, created in the frame of this project, is also open-source and available on our Geo4All Lab’s GitHub page.

In addition, our study focuses on segmenting point clouds in an almost fully automated way. The processing starts off by a cloured point cloud which is generated by ODM from nadir and oblique images. Then, a Normalized Digital Surface Model (nDSM) is generated. The Cloth Simulation Filter (CSF) algorithm is used to separate points on the ground and a Digital Elevation Model is generated from those points. From the nDSM the ground and the near ground points are removed, this way the low vegetationare also filtered out.

Subsequently the filtered point cloud is voxelized. Voxels are essential to divide the complex task into small processes which can be parallelized. With the help of a sequential RANdom SAmple Consensus (RANSAC) method in each voxel, one or more significant planes are detected. Those points in a voxel that fit on a found plane are substituted by a single point on that plane and the normal of the plane, thus a spare point could be used later. This way noise and vegetation is filtered out in a robust and efficient way.

In the next step of processing the spare point cloud is segmented by the normal directions into three categories: walls, roofs and others. The wall and roof points are further segmented separately by region enlargement method. Finally, the continuous wall and roof segments are combined to define the footprints of the buildings.

Test areas and traditional land surveying methods were used to validate the aforementioned algorithms. As our intention is to apply the technology mainly for smaller settlements, we are about to focus on detecting detached houses. According to our preliminary results, land registry maps with homogenous accuracy is achievable. Accuracy can be characterized by less than 10 cm, which meets the requirements in general. With the contribution of open-source solutions, the technology offers an economical way of updating old and heterogenous land registry maps.