Disertací k diskreditaci... 4
Further research
In general, my research led to several key findings, first I tried to find the best ground filter, because in our complex study area there was a need for more detailed data than are commonly available. This led to use point clouds not only from different sources, but also from different points in time. All these findings were successfully achieved in our scientific project and were subsequently used in further outputs and studies. However, there are a lot of unresolved issues and a lot of interesting applications in which ground filters are newly tested (e.g. ground filtering of raster satellite data, UAV LiDAR data, ensemble filtering) and I would like to use the experience gained in this thesis to further explore these possibilities.
Currently, UAV LiDAR (fusion of LiDAR sensor with the UAV platform) is becoming more available and represents one of these possibilities. In contrast to ALS it provides denser point clouds thus brings new challenges for ground filtration because of lot of noise and hardly to reach symmetric flight lines. On the other hand, there is great opportunity for very accurate estimating ground elevation and vegetation characteristics.
There are indications of the use of ground filtering algorithms in satellite data processing, e.g. filtering of DSMs from TanDEM-X satellite. Most satellite data (as well TanDEM-X) include canopy and building measurements because (same as photogrammetry) radar waves cannot penetrate fully through vegetation and buildings to reach the ground. Therefore, filtering methods are needed if DSMs from satellites are to be used further. This approach is quite challenging mainly because there is no filtering of point clouds but raster data. So far, few filtering algorithms have been tested (e.g. morphological filter) and it would be worth trying other filters or completely new methods, such as ensemble ground filtering.
The ensembled filtering method can divide data (points/pixels) into problematic and non-problematic parts. This can be achieved by using several different filters in the same area, where most algorithms agree is not a problematic part and vice versa. However, if the results differ, it is necessary to identify the part as problematic and then focus on it. Ensemble filtering does not necessarily use several algorithms, but also several different parameter settings within one algorithm. I would like to focus on this method firstly in my further research.
One of the other directions of further research may be various data fusions for which it is necessary to filter the ground points. Fusion of RGB imagery information with UAV photogrammetry-based point cloud or with LiDAR point cloud constitute an interesting data source for filtering. It is questionable if the additional RGB information could help to better detect ground point. A very similar principle can be imagined when filtering point clouds within states where the parameters of filtering can be adjusted using the land cover type (e.g. Corine Land Cover).
Plenty of filtering algorithms have been presented to distinguish ground and non-ground points from point clouds. However, with the existing methods, it is difficult to derive satisfactory filtering results. The creation of a completely new own ground filter of point clouds, which would, for example, adapt itself to the surrounding environment, can be considered a very distant future research. Some attempts have already emerged, for example with a slope estimate, but mostly without much success. In my personal opinion, the usage of neural networks and machine learning for ground filtering seems very promising.