Hi, in this episode of the vlog from a process of creating
our independent game "End of the Sun" we show you how
the whole process of implementing
a huge 10-metre rock
into a computer game
using the photogrammetry.
So out we go – to the "Stone City" (polish: Skamieniałe Miasto)
We are once again in Skamieniałe Miasto in Ciężkowice
Skamieniałe Miasto in Ciężkowice, where there are many rock formations from sandstone,
remainings of
old quarries.
They form some interesting shapes.
Okay, and here we've got shots
from the drones and camera
of this „Witch" rock
that's how it looks.
And now it's about
getting each photo to contain 30% of the information
from the previous one.
What does it mean?
If we have this picture, then another, the program must find
the corresponding points that are common
thanks to that it calculates their offset and on the basis of that
it calculates the so called 'Point cloud'.
Two of them are being created
one is quite thin,
it looks like that.
This is the first stage
the program adjusts these photos and puts them in the appropriate positions.
Oh, here we can clearly see in the picture
in this mode
that each of these squares
is the position from which each photo was taken.
I pick a photo, and it highlights
that this photo is taken from here.
Here we have all of the pictures taken with Lumix
from very close and it knows
exactly where the photo was taken.
Some photos based on
others determine their position
After this thin cloud of points
we prepare a cloud of thick points
and it looks much better.
You can see all the details,
but it's not a model yet.
This is also a cloud of points
as we come closer, it is evident that they are spreading
there are not there.
After this stage
this stage
we create a model
just before we do it, as we can see some elements
that we did not want appear, trees etc.
we have to remove it all.
We do this so that we
select our thing
and highlight the part that we want to remove.
And this is some tedious work,
but thanks to that only
what we are interested in remains.
And only when we do it on this basis
will we calculate the model, and then we will prepare
the texture. What's even more important
when taking pictures with multiple cameras?
One of them was a drone camera
that had to catch a rock from the top because
it was impossible to do from ground level.
This rock is so many meters high
that you had to fly from the top of the drone
and without this it would not be possible
using tripod or something like that
This is the most convenient solution,
except that the drone camera's focal length is different
than the focal length of the camera
from ground position.
You need to adjust
these
focal lengths manually on some pictures
but this usually happens automatically. This angle is
narrower
than this and therefore we need information
so the program can .
take that into account
Now we have prepared our thick cloud of points
without unnecessary elements
it looks like this.
Very nice, also from underneath
all details are visible.
We will set a frame now - what will it process?
If there are any unused points
they will not be processed.
This gives us some
computational time
because for such photogrammetry
we need really good hardware.
Ideally some GeForce graphics with
the CUDA system to support it
and a very good processor, a lot of RAM
we work on 16gb, (second PC has 32GB)
not a great amount,
but you have to have =
some so it won't crash (Software).
Now on the basis of this point cloud we finally
calculated the model
it was a longer period of time.
As you can see, the model looks much nicer.
It turned out really nice.
We will see this from that perspective,
as we finish the model we will put it into the engine
and I will be able to show how it will look
from the player's perspective.
This model has as many
as 2 158 000 vertices
and it's too much to carry
for standard pc.
We need to optimize that.
I will make a retopology, I will simplify it
so that from 2mln triangles
it will go down
to 5-10k triangles.
This is a really big area,
the player will have to come close
so it cannot be blocky.
The next step was to
bake the texture and
we prepared it
basing on these images from various positions.
Normally the whole
rock was this size
I split the model in half and baked it separately
into two parts. Each of these textures
has 8192x8192 pixels.
The point was not to use a bigger texture
because it is very
hard to process for computers.
Therefore, to maintain detail, with great close-ups,
and at the same time to make the
computer able to carry it
I divided the model in half, one half has one texture,
and the second - the other one, but everything fits later on.
This high-profile model,
so-called high poly,
we did his retopology,
and it finally looks like that.
As you can see, there are not too many details in here
but using the free program called XNormals
by putting low-detailed and
high-detailed models, this program
calculates the so-called Normal map
and thanks to that
computer finds it easier to process.
Typical technology used in games.
This gives us more details
on a very low detailed model.
We have picked some
spring flowers here
and we will take some single photos of them
We do it in this way,
on the sky,
and then in Photoshop it is easy to cut out.
It will be on a single polygon
(surface)
o in the game you will not see this space
but it is concerned
with optimization issues.
We cannot model this flower and put it in game
because no computer is able to process thousand flowers like that.
So a single image is taken and put
on one polygon.
This is not the final view yet, because this rock will still be refined
We have such details here - for example, here we baked
a piece of a tree.
And that's because the program could not know what was behind it
because it was presented in every photo.
So you have to manually paint everything. But it looks more or less like this
so it is already very good.
This is a texture of 8000x8000,
however, only 4000x4000 is on right now.
As you can see, we still
have a lot of details here.
Now let's see what gives us
this normal map.
We will turn off the basic texture...
Here we have a normal map overlay,
and I'll turn it off and on. Now it's on
and now it's off.
This is the same model - it has about
7k vertices,
and yet after adding a map we get as much detail.
This is how it works - the program calculates
such a normal map basing on
a highly detailed model
and it looks like this.
Here are all the details,
so the computer makes less calculations
and at the same time it
looks very good.
In eyes of the player it will look very cool
because here the perspective changes completely.
You see the size and details of this rock
they give the cool effect that this is
a mighty rock, not something tiny.
We do not have a collision model here,
so we can go through it,
but it's all going to be done soon.
Another interesting feature in Unity 3d
it has a very cool texture optimization solution.
This of course is also available
later from the game menu.
For example, we now have the best quality, that is,
we set the highest resolution of the texture
that is, say 4k.
Let's see now how it would
looked at the
lowest settings.
This is a 1/4 texture,
and it's 1/8
you can see the difference.
Now I switch from 1/8, here is the resolution of 1024 on such a large scale
and this is full res of texture
I wanted to show the rock we had created once more
and compare it
to the one we have.
You can see what a big model it is
which is differentiated everwhere.
There is not one repetitive place
and this is a great advantage of the
photogrammetry.
Không có nhận xét nào:
Đăng nhận xét