Mapping relief elements from laser scanner point clouds
In the base registry of the Dutch topographic maps (BRT, map scale 1:10.000), relief elements are objects that represent height differences in the landscape. Examples of relief elements include embankments, steep natural edges, and so-called Walls and Earth Walls. These elements are recorded as line objects in the database.
These objects are challenging to map automatically. The Dutch Kadaster is looking for opportunities to increase the quality of automation in mapping relief elements. In the Netherlands we do have a highly detailed height model, captured by airborne laser scanner data, which potentially can be used in deep learning algorithms to detect and map those objects.
That is exactly what this MSc topic is about.
What are the capabilities of the state-of-the-art deep learning networks to detect such an object in laser scanner point clouds?
In a literature review, find the best possible networks as a starting point;
generate training data and adapt the network to achieve an optimal result for detection
Optionally: detection is one step, mapping is another challenge. Depending on the state of the art, mapping may be included in the research.
Include user perspectives to understand what is the optimum between having too many detected elements or too few.