well thank you uh so well minnesota for is um
i would percent might work that has the title information gain view planning for
free form object tracking destruction without really time of flight cameras this work has been
uh and on in collaboration with the german interspersed centre and you'd of politics or
dlr
with uh segment real and that some folks and might supervisors in america
um
okay the presentation is divided by the motivation of this plane
what motivated to do this work and then i will just possible to the main
algorithms that and how they may not way that works that's more or less with
the that decision was trained how the information gain representation works how we compute the
different kind of a view one generation
and what secretary on that we use in order to choose those viewpoints that i
will show my results and i finally conclude with my conclusion
so the motivation is active view planning and why did we know um in cs
given an unknown seen what we try to do is to move our sensor in
space in order to get more information and get more data of the of the
scene in order to
a build a model
so i went i'll uh how objective is to do this autonomously a two models
three uh an object in three D this object can be uh of preformed it
doesn't have to be of any form
and i one of our prerequisites is that it has we don't have any kind
of information of the see that different methods in the literature that the use a
three model based or some kind of course in order to get oriented uh through
the so the modeling of they opt
and our proposal is to use the information gain in order to decide which of
beers we are going to use in order to uh build our model
so
our main algorithm it's looks like this and it's embedded in mainly in four steps
and the first one it's the data acquisition we use a three D time of
my camera in order to get a point cloud from the image so once we
extract this images the for the second step is still update um some internal in
representations
uh_huh one the principle one is an occupancy grid it some of the resolution not
occupancy grid where the data of a time of flight camera get stored not only
like the point cloud but also the you statistically um and so trinity of those
points
i will explain it one of these steps like product so it gets good so
after D is that this firstly what we get is we have data now the
representation and its image representation from this mission we can compute uh like the boundaries
of these a rematch uh in order to select some pubes that a formal the
bikini of that match
so once we we've got these views what the out the main part of this
algorithm that's it's to decide between these views which one we should choose in order
to get more information from the model okay and that's nonviolent decision maker and it
gets information from the uh mesh representation of from the occupancy grid that is the
one that has all the inserting at of the model okay so now i will
go well we show all the steps of the know to quite high and then
i go
and explain uh each step called so this is that the same um first be
that we saw before so we've got an initial posted we set up yeah like
anywhere but just looking at the scene the only prerequisite is that it's looking at
the scene then we get a beer we have they the both um representations and
then we simulate a those be used in our occupancy grid in order to get
which once we are supposed to have this uh information gain once we have one
that it provides the highest information gain what we do is we go with the
wrong what we choose that goes and then we extract another point from the set
and again this is done repeatedly until the algorithm finish and completes the model features
just before uh it changes to the presentation of a mesh tries uh
we could also be to um be used can be a providing more information and
then it computes again information gain of those be used and then we select one
in order to continue modeling them uh the object
so the first that it's the data acquisition as i already said we used uh
a time of like camera in this experiment we were using uh the message imaging
asr a four thousand
it has to be said that it has been calibrated and characterize what we use
but we signal you rate it is not only the intrinsic parameters as normal parameters
do but we already calibrated F measurements
by amplitude done by all kind of errors that these cameras
so but even the when we finish this calibration this camera are one of the
disadvantages this comment is that they are still have noise in definition and so i
think that what we do is characterized that so each pixel has a covariance associated
with depending on the definite it's mess so it's pixel has that's really covariance
i related to it
so once well for those who doesn't know about this time of flight cameras they
provide intensity images and that uh images over just a correspondence or one by one
and they are rolling resolutions like one hundred seventy six or one hundred forty four
um pixels but they provided that uh twenty five frames per sex of a up
to approximate so they are very fast enough to get this just
so once we have uh
and this camera appointed to always seen what we do is to get a point
cloud and this point cloud gets updated you know an occupancy grid this occupancy grid
is some of the resolution occupancy grid
and the first two
and this occupancy grid it's first field in with nothing and feel nothing wouldn't we
understand like an unknown area and it's just one box with a high and something
to do then as far and as far as we are getting um introducing point
close to the occupancy grid all these um pixels in space this box and get
updated with new a measurement and these new measurements modify the entertain it is inside
all these boxes
okay so we've got an example of how to measurements like you pathetic a mess
of measurements will be if they were like ninety degrees
of each other that would be a box so the first row with the but
so without any kind of information and then we've got before the updated will be
these two measurement is to covariances and after updating the model like for using the
entertain it is it will get something like this okay so that's the formulation the
typical information gain at the
so this is only for you know uh putting the answer to anything inside a
model and keeping it so after this that what we in this produces its it
give us you know sensor directionality so each voxel stores that covariance in the direction
if the covariance hasn't direction and usually does that value of the measurement has a
higher and today needed and not the X and Y values so it gets a
story in each voxel in which uh direction is it has been taken and the
good thing is that
this allows to model refinement so at the end we can choose which be used
we will be able to choose which we use you know um give us less
um even more information or reduces morgan's identity of certain areas
so once we update this these representation what we do is to create a uh
a match
in order to get more candidates use to uh to check what's uh information gain
we provide so these candidate viewpoint generation is made on a more uh of gradient
that he presented in like the two thousand eleven
and what it does is it builds a at an alarm age it detects boundaries
of this image uh given certain parameters like the length of the of the boundary
or do you deviation of the comforter of the of these boundary then it separates
them and then what it does it grows uh region inside this match in order
to fit a quadratic patch
in order to so this but i think but
alright
so it's fitting next to the to the previous iteration in order to uh i
sure some overlapping between the two beams and then send you be it's extracted from
this from each button okay so after D is what we do is these new
bills we simulate an in the occupancy grid and then we take the information so
how is done in X
slide
so what we do is now that we've got these deals that we extracted from
the viewpoint planner we come back to the occupancy grid and then we take like
assimilating those of use a C as if we were extensive so we ray tracing
with a ray tracing in order to see all in which areas would collide our
readings and C of those readings how uh the information gain will be okay so
for each one of these like point one of the point clouds simulated pointless we
start the covariance we do this using the same pushing that we did it as
we needed in real and then we compute the information gain based on this formulation
so what it does it's just like estimation of all the logarithm softly traced metrics
that it contains all the updated um
covariance matrix
okay so by doing this we're piddly and at the end we manage to get
our results and these are the results of we
we obtain by three we tested by a on three statues with different shapes free
form
yeah as you can see we get quite a very nice property um models of
then you can see some areas that they have been not feeling or model but
it's you to the configuration because the studies where on top of uh like a
little
chair and then the robot can not access to certain band
and you can see that they are not there we define uh models in some
of them but that's mainly because of the resolution of the camera it doesn't have
more resolution
so and for concluding i presented uh this new three D information gain new method
for viewpoint selection
um you to its internal representation its simplicity allows D model refinement so what in
the future we would like to do is to define liking which resolution we would
like to have a model like or in which sent which parts of the model
we would like to have more resolution so in order to try to get uh
a better a better model so we could even decide like by if it has
a lot of curvature that's an interesting place so we would be able to get
more refinement of these in this i
that's so thank you
something how
i
no
not for about four oh i
not like not one thousand times but i cannot guarantee in a certain like number
of time
no it doesn't really concludes by construction it will and it's like definitely because it
will always fit anywhere we have that are no and then at some point it
will you know like calls the object
but it at like in this in this one we had to close of like
manually because we what we had restricted the area of down because we could not
go down so i can not shown in simulation we could we could do everything
like
but i can assure a number of leaves i can assure that they will be
close to a minimum because it's always
by construction it's obvious
building it incrementally
sorry
how this
i
so it's quite a D
uh i yeah actually yes
so
there's a distance like the got the camera has its calibrated that sent a thirty
centimetres so you can not move far away from the object always in the in
the distance that you probably because they are quite sensitive in that and yeah with
but uh i'd like what we assume it's like in this overlapping it has to
be like a at twenty percent of the of the of the first row of
the of the camera and
yeah and then it follows the angle of the of the product fitting surface
sorry
yeah well with that
it's the ones that fit
no
no
oh
like
by construction so in order to refine the
the model you will be like getting new be used from different places following the
same structure because like usually what we have like they may never it's in this
at fourteen is what is plane and then you'll see a in a like you
matching just one point and you've got
see
i can be structure will be like the nicest when i just put it like
you know also normal way you just do a reading orthonormal way then you will
get rid use your and your covariance as much as possible but rather than these
i will not be able to get like if you decisions in which and i
will not get better than this like this is the best refinement that i can
get or calibrated camera better in order to get of reviews like this
yeah well yeah that will be able consider that
so what do i actually probably
okay i just a method of us that some folks and what he does it
calibrateds so this cameras um like they have ever skin distance for certain in distance
for each certain distance they have an offset the different often it follows a signal
to dial uh um function so you can get uh
you can then the detected and use it we usually calibrated sorry it's all in
the process calibration is like with a normal battery like the one that we use
in
in four intrinsic calibration but a huge one so that just the huge went and
then usually we use are different gray scales in the in the button so we
can because i'm different amplitudes the camera reacts differently so we have a different number
of incomplete you seen that depending on the intrinsic uh on the integration time that
we choose so all these parameters have to be chosen like in this experiment was
chosen for thirty centimetres and you calibrate the camera for that like for a range
of these
and then with this that with this pattern what we do is like we can
we compute all these uh functions that uh minimize the ever by uh we projecting
a plane like with the usual optical uh weight so what we do is you
know you can get because the intrinsic a parameters and then you were we like
put the plane on the space and then you mention what the mention that you
can get
for
i don't know it get it right
yeah