Hello i'm Marc Degroseilliers from technical
support here at Pix4D and today i
will talk to you about quality
assessment. Now this is a very important
topic and we get a lot of requests about
this and the question is how can I make
sure that my project looks good and also
is accurate in Pix4Dmapper.
So we will go over four different methods to make
sure that your reconstruction is good.
The first one is visual inspection and
that just means you generate the first
step of your project and then you
navigate the point cloud a little bit
you make sure that the features are
where you would expect them to be that
your model is not curved down or curved
up for example in the image that you see
here if you look closely you will see
that the buildings are slanted so
clearly something went wrong and so
you need to look into it and
fix it to make sure that you have a good
project so it could be because if the
imagery or it could be because of the
camera parameters but these are just two
examples. I said that you would need
to process the first step but sometimes
it's a little bit hard to tell and what
you can do is you can also densify the
point cloud in the second step and you
can see here on these images on the left
it's kind of hard to see what's
happening while on the right you can see
that there are patches which are not
correctly aligned
presumably this field is flat and so you
have you have some issues in this
particular project.
The second way to assess the accuracy
and maybe the first tool
that you want to use is the quality
report and you want to analyze this
quality report and make sure that
everything looks good and this is the
first table that is under quality report
that you can see here
it will talk about the number of key
points per images so for example in a
typical RGB project would like this
number to be at least 10,000 and it will
also talk about the number of matches
things like this. So as you can see on
the right you have these green check
marks and here if you have a red
checkmark then this is a red flag and
what you want to do is you want to
inspect this aspect of your project.
One thing that could be helpful is we have a
nice support article which is called
quality report help so it will help you
troubleshoot each and every one of these
sections. Maybe one last comment is here
you can see you have a yellow checkmark
in the in the last row and that is
because no 3D GCPs have been used but of
course you can get very good projects
without 3D GCPs so if you don't have to
3D GCPs and you shouldn't worry too much
about this yellow checkmark however if
it's red
maybe you want to have a look at this
quality report help support page.
We have a few graphics in the quality report
that help you get a quick overview of
your project and maybe some regions
where you might have problems
the first one is this graph here where
the blue dots are the original camera
positions from the gps data and the
green dots are the optimized camera positions
and as you can see in this graphic they
match very well so that's that's a good
sign that tells you: Ok Great! These two
pieces of data fit together well so
that's a good indicator that
your reconstruction is good
what you don't want to see is something
like this where the red dots represent
uncalibrated the cameras and
the difference between the blue dots
and the green dots is a little bit all
over the place in all sorts of different directions
that's not that's not a good sign
There's another graphic which will help
you estimate the overlap as you can see
here you have your whole project and
then for each region if at least five
images can see this region then it will
be colored green and nicely homogeneous
green is what you're looking for for
your project.
So for example this is a project where I
would have a little bit less confidence
in. So at the top as you can see you can
see quite some yellow and red and so I
would be less confident that
reconstruction and the top part of the
project is good and I would inspect it
more closely.
We also have the 2D keypoint matches graphic here every
image is going to be a vertex and we're
going to draw a line between two images
that share a key point so what you want
to have is a nicely connected graph
which is quite dense and you can see
here there's a lot of blacks that means
that there are a lot of edges
this is quite good. What you don't want
is something like this here where you
have several blocks. The blocks are
color-coded and here you can see
there are not enough matches between the
different images and there are some
blocks that are independent and then
they could be oriented
in the incorrect directions.
One of the new features of version 2.2 of Pix4Dmapper
is the addition of uncertainty ellipses
now what this will do is that it will
draw an ellipse around each camera and
the uncertainty is the size of the loop
so that bigger ellipse will mean that the
software is less sure about this
particular camera position and
orientation. One important thing to keep
in mind is that these ellipses they have
a magnification factor which is
displayed below the graphic so it could
be that these ellipses they are
magnified quite a lot and so you have
some ellipses that are bigger but if
they've been magnified by a large factor
and this doesn't mean that your project
is not good
it just means that the uncertainty is
higher in the regions where you have
bigger ellipses so then what would
recommend is to inspect this region and
make sure that the reconstruction in
this region is good so it gives you some
kind of areas to target for
inspection and how are you going to do
this well this is the third way that I
want to discuss and its projections in
the raycloud
so what you can do is you navigate in
your project you click a three-point and
in the right sidebar you will have all
the images that project onto this 3d
point and there will be a green cross
that will correspond to the pixel that
project onto this this feature that you
clicked in the raycloud and what you
want to have is you want to have this
green cross always pointing to the same
feature as in the images that as in the
image that you see right now. Here is
an example of something that didn't go
so well you can see that the green mark
is on different features in the
different images and so then that
tells you that you want to
fix your project, something went wrong.
In this particular case it was the
camera parameters that were not adjusted properly.
The last thing that I want to
mention are checkpoints. Checkpoints are
very similar to ground control point in
that the they are features in your in
your project that have precise
geolocation data that has been
captured while you were flying your
project for example. Now you've got your
location data for for this you're going
to click it in the images much like you
do for a GCP, the difference is that the
software will not use this information
to reconstruct the model what this means
is that when you have a GCP it will help
the reconstruction of the model because
the the software will try to fit the the
data the gps data to the point where the
3D GCP is. So it will help the
reconstruction. On the other hand
when you have checkpoints the mapper
will ignore this information and we'll
do the reconstruction as if it didn't
have this information. Now what's the
point, you might ask? Well what it will
provide you is an unbiased estimate of
the accuracy because the mapper didn't
try to fit the point it just ignored it
until there's no bias involved.
Ok that's all i wanted to say about
quality assessment today. I hope you
found this helpful and i'll see you in
the next video
For more infomation >> Heart Touching voice of a Young Boy - Chitta Chola By Jahangir - Duration: 3:47. 
Không có nhận xét nào:
Đăng nhận xét