AHN (National LiDAR) and Imagery Conference.

Last Wednesday, GeoSignum gave a presentation at the AHN and Imagery Conference.

A wonderful, informative day at a beautiful location, filled with conversations about elevation data, imagery, and the future of 3D information.

One central question was at the heart of our presentation: How do you go from a point cloud that you can visually understand as a human, to information that is automatically usable in management, analysis, or a Digital Twin?

A point cloud can look beautiful. But it is not something you can directly plug into your management system. The value is only created when you can convert the point cloud into usable information. For example, in the form of vectorization or 3D models. And preferably with the help of algorithms to minimize time-consuming manual work as much as possible.

To do that, an algorithm first needs to understand what the points represent.

What is ground level?

What is vegetation?

What is a building?

What is a pole, tree, facade, or noise?

AI can significantly accelerate this process, but AI is not a magic button.

In practice, we run into all sorts of edge cases: datasets with too much noise, too few points, objects touching each other, or objects partially blocked by other objects.

Because ultimately, you don't want the algorithm to only work well on an ideal dataset. You want it to remain reliable in practice: dealing with noise, overlap, occlusion, and variation in the data.