Hello everyone and welcome back to computer vision lecture series.

This is lecture 8 part 1.

In this lecture we are going to talk about dense motion estimation.

These slides are based on these references.

You can look up them if you want to take a deeper look into this topic dense motion estimation.

But otherwise for the purposes of our course and the content that we are covering this

slide is enough.

What do we define?

What do we mean by motion estimation?

In motion estimation we have to determine computationally the flow of our object of

interest.

What does dense mean?

Dense means that for each and every pixel we have to find this flow.

And usually it is done in a video or it could be a small animation or a gif where there

are at least two different frames to compute this flow from.

So where does this flow comes from?

Usually the flow or the motion that comes from either the object moves or the camera

moves.

So if you are changing the position or location of your camera then there is an apparent motion.

Or if the object itself moves when the camera is steady.

In this case for example on the left hand side you see a car here and this is a flow

field or the flow generated by this car in a small video.

How do we determine this flow?

It is through some vectors which show you the strength and the direction of this flow

around this object.

And this kind of flow field generated flow field is called motion estimation or a flow

of this object.

Basically we are trying to quantify the object motion computationally and that is what we

call as flow.

Okay and another reason where the flow can also come from is camera movement.

So if your image is stationary and if you generated another image just by moving your

camera then the flow can result from that as well.

By estimating this flow we can estimate the magnitude or the relative depth of the objects

inside the image and therefore flow estimation or motion estimation is useful for depth estimation

as well.

As in this case we can see the flow field here is generated and you can see that the

higher or the more stronger field are visible in the trunk of this tree whereas lower or

very less magnitude fields are seen around the sky areas or objects which are farther

away from the camera.

And this is due to the perspective view of the camera because when you translate or transform

3D object into 2D image plane the objects which are closer to the image plane will show

a higher magnitude of flow or motion estimation or motion fields and objects which are farther

away will show lower values of motion fields typically.

Basically as you see here the dense estimate of all the pixel values is considered as dense

and it is not only for one particular feature like the trunk of the tree or the house but

it is actually for every pixel in the image.

There are different applications for generic motion estimation as well as dense motion

estimation.

We will see some examples now of what it means to have this kind of flow or trying to compute

this flow.