Hello everyone and welcome back to computer vision lecture series.

This is lecture 10 part 1.

In this lecture we are going to talk about structured light.

Until now in the past weeks we have seen how we can calculate depth from stereo vision.

And specifically using dense motion estimation techniques like optical flow and stereo correspondences

we could find the differences between two images or sequence of images and localize

every pixel in 3D real world.

And the location of this pixel in 3D real world essentially gives us the depth of the

pixel.

So what is structured light?

Until now we have seen in stereo vision the image acquisition techniques.

It is called passive image acquisition technique.

Why?

Because in this case we are just capturing images.

We have a setup of stereo cameras which are separated by a fixed baseline.

We have two different images or more images and then we reconstruct or recreate the original

scene from the two images or multiple images.

So in binocular stereo vision or in dense motion estimation what we are doing is using

multiple set of cameras to reconstruct the 3D depth or 3D locations of each and every

pixel point.

So the reconstructed part of the image gives us the depth also of every pixel in the image.

This is called passive image acquisition.

In this case the correspondences sometimes fail when the feature distinctiveness is not

very high or we are not able to find nice correspondences.

So there are issues as well during finding stereo correspondences.

We have seen some of those issues.

Mainly it was about missing distinctive features.

For example when there is a plain surface it is not easy to find very nice feature

locations or feature correspondences in different images.

And these are the places where the correspondences fail.

And therefore there are some missing parts here which we are not able to recreate because

of those missing correspondences.

But in general it gives us a good depth map of the original image.

So today we are going to look into what is active image acquisition.

In active image acquisition what we do is we insert an additional source of light or

additional patterns.

For example on this left image you see this pattern of light which is projected onto this

sculpture and we capture the geometry of this sculpture using the information of this pattern.

What we do is when we project this light on this sculpture we already know the geometry

of this light pattern.

So we are in control of creating the kind of pattern that we want to project as well

as the location of this projector.

And therefore using this same geometry or the same setup as stereo correspondence we

can recreate or we can reconstruct the shape of this object by this method.

So what we are doing is repurposing our stereo correspondence algorithms by additional input

like this pattern and using it to create a better more dense correspondence or more robust

correspondences to be precise.

In the middle of the image there is a nice example of this setup where in the 2000s the

sculpture of David by Michael Angelo was given access to some scientists from US and they

were able to recreate a 3D structure of this sculpture using this structured light algorithm.