Bestand wählen


Zitierlink des Filmsegments
Embed Code

Automatisierte Medienanalyse

Erkannte Entitäten
in 1 of his problems around given were was was a model will later be myself employer gets it is faces
shiny objects so I might my problem this is actually similar
i i I i i hate sometimes others that don't yet exist for reasons I don't know sort of has no solutions looking for a problem and at the prep for the city's search
for much get close the subpoenas fought on
end sometimes makes the do silly
things I didn't do this but I was
so bear with me and talk about uh solution to a problem that didn't exist until recently and that's that's open dramatic
I'm going back in time about as a connections must be yeah so going back in time and the and the year so of
beckoned 2004 allows working
on a common on PhD in forced ecology man and the excitatory still seasons and build up on but that would come from so forget to do while I was there was a get to fly a big balloon as liquor axis most a skillet there was a the key was a low nominal so because 11 foot hello left on which the FAA restrictions on link to the balloon payment but don't don't violent foot balloons OK 60 that's as big any on unless you wanna get it the effort involved on but show that that that never happened and so a few large bloom
with an array of cameras I took a bunch of with bunch of images from the rate
had that had some modified cameras others nicer Kodak that DCs on us as it was beautiful camera is a pro sports camera so I had amazing ISO levels and unlike now it wasn't it wasn't some filtering process was sort of smoothed and and moved the pixels are officially no they had that each pixel was had no extra work electronics associated with that so the whole pixel was the the actual past itself so although like hitting that it's since the full frame is a full frame of CCD all like getting that got back and so here it is Tuesday I think those made faction 2001 2002 it was able to to do I those of 40 890 600 so really really fast on the camera for the time of also I bought a few years after Kmart enhancement 14 thousand dollars a on certain but some but summer of filters in front of it so near Fred filter and its and its band-pass filters on those band-pass filters were were not biological research 1 understand OK what's happening with leaves at leaf level the light stress so lots of fun stuff on the total failure of a project way too early from because there is no trivial way to figure out where we were taking pictures of and so several can computer vision outside about and then realize that neither had computers nor the code beginning with the some of them some few years ago I saw condition of because she noted that the body give them some geo conference of some sort and is talking about the future of G or many amendments action is probably uh but whatever was so that future geo and wearable devices and always on cameras and reconstruction of 3 D space and in a sort of all all things that that so that they could come to be and started getting it's a sort of was a worm in my head it's reminded me of the problem that could solve for which was how to take these arbitrary photos that come from man on the come from whatever on the main I have GPS associate with them and construct them into something that we can use in geospatial world and so this which over the presentation here for open drawn map the
the so the idea here
is somewhat of a small the name of the but it's not really a drone so it's not really a map and for those of you who really like the esteem Apache and and and all the relatively liberal licenses of its open like copyleft so
open don't matter to amend project on so the idea is that we go from those arbitrary images through some server processing pipeline into point clouds Digital Surface Models textured that digital surface models so this is like a full immersive 3 D environment and ortho injury plot 50 stuff and you know than then cost directories the hands and classic in Rasta surface models and and so then the question becomes how old you there and so by watching 1st question what is it and so stressed it's possible changes US injured visible on based on the 3 basic technologies in education within stand for so long because of all Bundler DS and on these are role in this is all computer vision code this is this is stuff that's this global reference will solve and thus means that outputs of which text files is the data a muscle to accept but you know it's really really cool stuff on there and it's really deployable underwent 2 2 0 . 0 4 3 2 bit compensates for that topic and 14 4 but will fix over and above the yeah were embarrassed so what is not secondly but it's it's actually it we there's function which this week on right now it's a place places a full 3 D of point cloud of what uh but it's not an searches that are more replicate things much loud and out here and as far all 3 point cloud mesh editing environments we can do on fancy cockatoo things on there's good tools for that and and much much less than that of the useful her views but there's a bazillion YouTube videos showing exact value so there's at least some some pedagogy for forgetting that and it's not like planning software and of them solve everything all also but it's not yet complete on the was so the uh bowl mash creation texture much creation an author photos and all the social things from but that's really close so long in the short term will be essentially single menu and command line and you by pure images through an image files and long term of the right that rapid web front and I would like to put put it easy sir Python wrapper and Q just for running that and causal terminal access and then data structure something that that will need to be addressed really pushes the PostGIS or some sort of a file-based uh uh this structures on so
I will will we allow our work said but we think it through to the sequence and this is this is done and apologized but the rest on so 1st the color photos and you do the full stereo matching the RGD pasta imagesetter basically distract here which ones are related to each other i gives you a sparse point cloud many do full stereo matching we compare every pixel to every possible pixel in every other possible image that's related and you dense point cloud and then you national so you turn on the
surface and then you start painting that surface with original images and the and and well maybe that's not the ideal way to do may we need this 1 then we need to do some some other us smoothing things me that look great but we can do that that's that's there that we we have the technology and that's pretty much it now you get extra service model now I'm going to connect studies in creating our sector products D to surface or on ranches classifiers to separate out the buildings for the trees around on and create the and such so let's let's do on
quick demonstration the which is the really tricky 1 by my the to really do like against someone of xt OK the
so model the the wonders aloud demonstration area but so
dense point cloud since this point cloud is constructed from these original images if we want the you
with those original images look like as the original image there's a dense point cloud this you can get now from open that and so the 2 much of overlapping images in and you'll get a nice dense point cloud and and and much like hearing demonstrated yesterday reported by yesterday from if you wanna keep your coffee warm all while you're running you will have that option also run on Amazon and
keep their server from swarm on so those those little original photos in his her see what's the notes look at a couple of on
the adjacent and again we're doing feature matching between them so this tree
here's the CM and uh
since this tree here and extracts a matching points on figure out how they are related and then we do inference magic where read it on some linear approximation of where the rays a sexy reconstruct the original camera positions and the original points don't make me go into a that little actually fully understand I just and which 1 sentence on so again we get we get a point cloud that's pretty nifty on now would be really nice will be
really nice this is a three-dimensional mesh so all and this need to national grams of and trees separately
from but there's there's a buildings there and mediation trees the the we get more effects moles that don't exist on I should say this is Seneca County Ohio and I grew up in northwest Ohio it's harder showed yourself letters on but that's real and and those teachers are about some somewhere between 8 and 12 feet deep so it is a sense of how much topography there's the rest landscape as well on so maybe next time will will do so will do more or less flat look at each location and so this minus machine but we don't know the rest actually done and national at for the time being on the and and OR kind in the script and the rest of the coop pulled and but essentially take those
photos images from this little texturing from register restorationist those the this is the the trait
back so now I'm going to do a from a high CPU process of life in front of the studio audience what happens I guarantee a computer crash the so what are we had we 1st the presentation do have sunk there is 1 1 here some then
dense I I'm not only someone's going to play the drums of data but so it worked it and single years
after bear with me so there's texture on this lecture and then in order to achieve that this is not full text and this this low resolution TextRank fearful resolution texture and be even prettier uvular see a satellite that road quite nicely on in full resolution but but pretty
snazzy would not texture back to model now the next steps will be and how we learn Northrop fire had we then geo-referenced but those the comparatively easy problems connection run that it just as that crash we're or time but but
it was simply precious and like saying yeah like dancing to the tyranny dramas in room yeah the 1 with guitar than the whole for the pressure this is all
GPU intensive and get ultrabooks so as to know a little bit more detail on a model using in and that the true do full
full texture this particular model would have an underground resolution about 2 inches so it's interesting things do that so I work for a of and the Banditron Park district and on on these reduce this is to basic products right this wasn't a fly in a in a number of natural areas including a wetland something of a district and identify I and this applied extents for restricting and monitor that over time with a really had a good way to sort of I really monitor the extents but also have had a good way to identify whether this tiny little patches of didn't die off right straighten out this spring Holger poison stuff dies you feel pretty good move on to the next place reinspect poison and 1 little plant to survives and with invasive plants that take over its mass and course you just killed everything so you know it's perfect between for that and that's what makes that's 1 1 application of this well and the other thing on do because we can get 1 inch imagery and color Fred and we will be doing now edification native communities so in this large well in complex it's very difficult to access a difficult to walk through on very difficult to map will have a sense of where some of us really interesting the communities are and potentially find some of the some of the things that we know we should be good stewards of and be paying attention to and and and the spring the heck out of all the data points and learning about so those those this and this sort of natural resource application the application of which is not as interesting and me from from the kind of project is the more interesting to me from sort of technical standpoint of the inventory of engineering structures that move over time engineering structures along the lake and this will be able to hopefully get some really good vertical accuracy and watch for changes under surveyors out to take individual points all of these things down and real 3 D space and then we can measure movement not just movement of points that are registered that we servers of to periodically that's useful we need them and but then also find out what's happening between but yes so the demo succeeded so no song yet I incidence relation GPS for us OK so there's another application here so this is drones open drawn map again I said it's not open as in BST it's not grown as and killing people it's not really a map from so it's almost as it's actually better named in OpenStreetMap because even last what it sounds like it's and that's that's what is in it for so let's let's go a little deeper here and then do something even less like I
drawn up the things all
1st of all so question distribution so highly highly then distributors how we show this in
and out of the dismal when the questions here but 1 of the things that will do is uh as pushes and push the ortho products from us and open aerial map and the couple the god this is the genesis of a couple other an idea and surface model of Nash type of projects track what global datasets of these there are there just certain Genesis stages and those will hopefully be repositories for all the data people process with this should it choose to but what's really cool is folks like I'm the I Howard Butler's end end and the whole should well OK I'm real names but Howard and invokes related to them by doing things like taking C + + stuff last zip compression libraries solider compression libraries and porting in Java script by compiling the by by using JavaScript is the Polish story that crazy stuff shouldn't be done has been done is wonderful on so now you can actually on through through a zipped it'll you last at each step a point cloud this visit as a as a as a class object and throw on Amazon Machine somewhere and actually viewing a browser had stream over as a few megabytes few hundred megabytes instead of the few gigabytes that it really is and then inside the browser decompresses and you can display it with web G it also so not only can you easily store this and and and give it to people on but they can easily view in use but a quick plug
from and and side at I want to foster G Korea a couple weeks ago of folks there wonderful next year it's in career you should go on and and you should be at least as of as silly as I was it's a very very very beautiful place on with beautiful people a a lot of energy and the geo side on and and and a lot of fun on the
reason I bring that up is swallows there has working with on something called literary who went on the query thing here that way so medullary
is for that for those who haven't seen it it's a it's intended to be a global crowd sourced from the dual street view of product l the route would pre liberally licensed as more lessons there will be compatible lessons with OpenStreetMap on and other duties of these wonderful things were a lot of way area so as my little street view from from
government God the campgrounds Korean every time I tried I was laughter very nice laughter but nonetheless there from from being detected strict in Korea and
N so that the series but you'll notice actually if you look at this if you look at this line it some I was not I did not have so due before I walked this site I want point this I was actually perfectly sober releases sober is an unordered racism is certainly had not consumed in any all Hall from then I think probably a week of what I wanted to this what and you can see just going all over the place because they had this this GPS-enabled camera that was not there good so far was a great but but that but that the GPS will migrate so he said well this is
a problem for open drum that naturally not have opened on that far enough along that time that she uses so I might have used a partial suffered patch this but we'll we'll talk about that I'm so here's a here's a subset of those points perfectly fine other than not being a straight line because of the extra walk along a sidewalk along road so if we take a minute you just just do a quick heat map we can see we that duplicate points this should be the should be just all on continuous red would this is like 1 point is like 2 points this is like at many points in any area basically every time it took a photo it wasn't properly thing the ephemeris from the GPS and so in addition of being wrong it was really wrong on so the that through that structure from motion
approach so we take each of those photos imagine to each other and match similar features in it in each and now we know the relative camera positions all along the sidewalk so now we've got not in real space not in not in accordance system that we know we get on that but we get some sense of what the spacing of those camera positions were so cool that's that's fine what we do with that well there go
there is there is sort of a geographically all runs at 45 degrees to the probably the models and optimizing it on the to to minimize the the the the x and the y axis so we do is we actually figure out what local courts are at the endpoints so look at this end point this point and do a quick go linear the linear equation to express that and now we've got actually georeferenced nifty only retrieve we see everything looks about the same it's even if we now take gone from this set of points to of well from the set of points to nicely spread out points and we shifted our geography is what and so if we can never on on the terrible cartographer let's be
honest on so here it is on top of a map on here is the original points nowhere near Tehran road and here they are in the correct location perfectly spaced and even with those little little places where the walk around groups of people are trees of as switching from the question so open dramatic and not just
runs but maps effect patterns not maps and openness and the GPL thank you on the thought Ch in the and the I'm
actually the from the ground standpoint thank you take care of the camera and just sort of go around and then dump out individual J. pegs we'll see and and take those overlapping images and the 3 D effect yes you can only you want a good lens so avoid yourself um but otherwise the at and low-resolution isn't necessarily a bad thing and so although like with with us was a 4 K camera in a retirement very low resolution images as well the each the video yeah yeah tasty should work their longevity someone's and from the other the other caveat is uh there may be some work so for for most sort of consumer cameras as an existing database of what the size of a scene last this and what what the what the focal distances and you know the relevant the relevant photographic parameters so maybe calibration process or something similar to look up on usually when I look up stuff and I don't know of and with video cameras but what they did to him as a library built into a other than that OK with that information was a lens from so if they it is cool so then you could actually add those all anyway now I know search search so that that that can plot of the train of thought the railed so yes so you can actually there's there's some and you where'd I condition that on the next next week at 2 on there's a place for then you can add those parameters to the to the calibration information in the in of dramatic and then right so thanks to the talk about a I'm curious about control of is it internal as that not part of the the part of the program so you mention senor surveyors of when they come back with some some control points there is they are you doing that very doing the the matching in a separate thing from open Ruiz control built and I have not yet so I as a starting place bypasses parameters and went from uh by the two-step process them around to get point cloud mesh etc. and the new and separate script to do the basic Hassan will start with just linear approximation model unitary transformation on apologist yet just asking transformations rotation scaling skewing on and then we can iterate from there from the user interface side absentees face but units at once the favorite thing on I don't know exactly know the work on the part of the reason for the you know a core conferences this is out of Libya so for pipeline and some people will tools around we will build some tools rounded as well obviously furious and also in the short term you can easily georeferenced point clouds and cloud compare partners another 1 of those you know that up to intuitive but great tutorials online sort of packages and it's not fair said well MeshLab or Cloud compare they do freaking everything in the world there may be no good user experience that can be created from something that does everything but and it was the question is and when you don't Fourier do-it-yourself infrared channel I haven't done anything yet but he was the infrared he had the cars hard find now on the other was the that the cameras made in 2001 and you still get it set I remembered it's a DSC 720 x the ACS because it really really fast you still get on ebay and if you're looking at it from the thousand actuations a few hanging out there the persistent search there I have not asked my wife of item by 1 of yeah a that's why search is still there 1 day 1 day you know maybe the will look for money and I think I can buy 1 when the public labs scattered some have seen their sizes curative you're doing there you know film and I'm totally I'm totally know by the uh 1 of the kids of divisions revise it gets last couple weeks and yes yes burned by the kids went from upon blooms we have authorization life and if they on for couple sites on but there's gonna be a lot of stuff we slide it it authorization think too long we just you know it's opportunity in 1 go out there so will drop a 6 foot 1 9 11 4 but a 6 foot 1 and 4 below on yeah actually with with with their cameras and by properly 3 3 4 4 4 4 4 4 ones thanks yeah public labs are so that's to really bring their going so much and
Quelle <Physik>
Offene Menge
Demo <Programm>
Google Street View
Radikal <Mathematik>
Skript <Programm>
Kette <Mathematik>
Metropolitan area network
Güte der Anpassung
Gebäude <Mathematik>
Wurm <Informatik>
Schwach besetzte Matrix
Dichte <Physik>
Dienst <Informatik>
Rechter Winkel
Grundsätze ordnungsmäßiger Datenverarbeitung
Ordnung <Mathematik>
Objekt <Kategorie>
Folge <Mathematik>
Klasse <Mathematik>
Räumliche Anordnung
Demoszene <Programmierung>
Virtuelle Maschine
Reelle Zahl
Orthogonale Funktionen
Automatische Handlungsplanung
Elektronische Publikation
Binder <Informatik>
Patch <Software>
Offene Menge
Maschinelles Sehen
Prozess <Physik>
Inferenz <Künstliche Intelligenz>
Natürliche Zahl
Kartesische Koordinaten
Computerunterstütztes Verfahren
Komplex <Algebra>
Einheit <Mathematik>
Maschinelles Sehen
Figurierte Zahl
Funktion <Mathematik>
Filter <Stochastik>
Prozess <Informatik>
Physikalischer Effekt
Klassische Physik
Plot <Graphische Darstellung>
Konfiguration <Informatik>
Projektive Ebene
Proxy Server
Web Site
Unitärer Operator
Transformation <Mathematik>
Kombinatorische Gruppentheorie
Digitale Photographie
Wrapper <Programmierung>
Bildgebendes Verfahren
Einfach zusammenhängender Raum
Matching <Graphentheorie>
Einfache Genauigkeit
Lineare Gleichung
Physikalisches System
Objekt <Kategorie>
Mapping <Computergraphik>


Formale Metadaten

Titel OpenDroneMap
Serientitel FOSS4G 2014 Portland
Autor Mather, Stephen
Lizenz CC-Namensnennung 3.0 Deutschland:
Sie dürfen das Werk bzw. den Inhalt zu jedem legalen Zweck nutzen, verändern und in unveränderter oder veränderter Form vervielfältigen, verbreiten und öffentlich zugänglich machen, sofern Sie den Namen des Autors/Rechteinhabers in der von ihm festgelegten Weise nennen.
DOI 10.5446/31744
Herausgeber FOSS4G, Open Source Geospatial Foundation (OSGeo)
Erscheinungsjahr 2014
Sprache Englisch
Produzent FOSS4G
Open Source Geospatial Foundation (OSGeo)
Produktionsjahr 2014
Produktionsort Portland, Oregon, United States of America

Inhaltliche Metadaten

Fachgebiet Informatik
Abstract Aimed at developers and end-users, this presentation will cover the current state-of-the-art of OpenDroneMap, toolkit of FOSS computer vision tools aiming to be easy to use for referencing unstructured photos into geography data (colorized point clouds, referenced photos, orthophotos, surface models and more), whether the images be sourced from street level photos, building interiors, or from sUASs (drones).Currently no such comprehensive FOSS toolkit exists and is easy to use and install. ODM aims to fill this gap.
Schlagwörter sUAS

Ähnliche Filme