LIDAR means laser imaging, detection, and ranging. LIDAR scanners are used in autonomous cars, in drones, in archeology and on filmsets. But the only thing they deliver are millions of points.
That is fine for drones and cars, which are using them to avoid crashing into things. But in many other applications, it is important what you have scanned. Do those points in space belong to the wall of a house or to the street? Is that a streetsign or a person?
A lot of research tries to answer these questions correctly using machine learning. We believe that that ambition is misguided, because you must train a model with thousands of LIDAR scans of every single thing that has a name - whether you want to call it a teaspoon or Taj Mahal. A much more reasonable ambition is to find the basic structure of things such as balls, boxes, surface, cylinders, cones or pyramids. When you find such structure in the millions of points that are a LIDAR scan, it is becomes much easier to search for concepts.
But first we must find those basic structures.
In the SINLAB, we have our VLP-16 LIDAR scanner with which you should create LIDAR scans for experimentation. We have also the open-source Blensor software that was extended to model all the small inaccuracies and errors that the VLP-16's scans have. You can create or load computer models in Blender and create scans from them exactly as the VLP-16 would using Blensor.
So you can develop, train and verify a machine learning model by creating a ground truth dataset with Blensor, then check how well it does with a VLP-16 scan of the real world.
Learning outcome
- You have hands-on experience with LIDAR scanning and rendering.
- You understand 3D data representation as point clouds and geometric models.
- You have insights into 3D modeling with Blender.
- You have practical experience with machine learning models that capture 3D structures ranging from regression analysis (RANSAC) to deep learning (GraphCNN or PointNet++).
Conditions
We expect that you:
- have been admitted to a master's program in MatNat@UiO - primarily PROSA
- take this as a long thesis
- will participate actively in the weekly SINLAB meetings
- are present in the lab and collaborate with other students and staff
- are interested in Python and C++ programming
- are interesting in 3D geometry
- include the course IN5060 in the study plan, unless you have already completed a course on classical (non-ML) data analysis