This audio is presented by the University of Erlangen-Nürnberg.

Big Bergschein.

Big Bergschein.

Okay, what else did we talk about?

Immediately it should come into your mind when you talk about the structure tensor, what was very similar?

The Wesselness filter, exactly.

The Wesselness filter, exactly.

Exactly, what else did we talk about?

After Wesselness everything becomes blurry.

We talked about image descriptors and multiresolution.

Oh, yes?

Yeah, exactly. Edge preserving filtering.

Which methods come into your mind when we look into edge preserving filtering?

Excuse me? Bilateral filter and guided filter.

These were the two methods we've been talking about.

And then we were talking about feature descriptors, right? SIFT and SURF.

I think we mainly talked about SIFT.

So let's put SIFT here.

And then we were talking about something super resolution, exactly, super resolution.

Okay, and I think this concludes our pre-processing.

And then we went to the next chapter.

And we will call this probably image analysis.

And the first thing we did is we did a short refresher course.

Ah, okay. We also have a couple of tools, right?

Tools. And the first tool we got to know was the singular value decomposition.

We will use that today again.

And another tool that we found incredibly, incredibly useful were projection matrices.

So projection matrices are really useful because we can use a matrix to define,

to describe a perspective projection in a homogeneous space, which was really useful.

And today we will use that to talk a bit about image analysis.

And the first thing we will talk about will be epipolar geometry.

So we will be talking about stereo camera vision systems today.

And one application where you can use that is, for example, magnetic navigation.

So far we've only talked about image pre-processing,

and we were essentially talking about a single image, and we did not use a combination of images.

Super resolution was a slight exception because we were using multiple images to compute one big image,

one highly resolved image.

But now I can use a couple of images and try to reconstruct a 3D structure.

So from two views I can reconstruct the 3D point if I have a calibrated projection matrix.

And in the following we will start processing multiple images, at least for the course here.

So one idea that is where you really need 3D information is magnetic navigation.

So let's say you have a system like this.

Can everybody see that? Well, should I turn off the front?

So this is already off.

Can everybody see that device?

So this device is for magnetic navigation.

And now imagine you have a patient on the table, and this patient has a catheter inside his body.

And now you have some X-ray projection guidance, and you're moving the catheter.

So we talked about this earlier, right?

So you're moving the catheter through the body, and then you can use this device to use magnetic forces