Bestand wählen
Merken

Compositing a Global Mosaic

Zitierlink des Filmsegments
Embed Code

Automatisierte Medienanalyse

Beta
Erkannte Entitäten
Sprachtranskript
it OK I think would have started 1 of so I'm Frank were damn I in 2 spatial developer of 20 years and PCI 19 nineties and worked as an independent consultant doing geologist the thing and probably best known for uh most of the 2 thousands and spent a couple years recently at Google and I've been at play for the last 13 months so I and you director for a long time although just finishing up my Sherman many more and the plane allows measure the data pipeline team so my particular responsibility is after we've got off the rectified images is turn them into most so plant labs is a small sadly company some so we build up operator Earth-observing satellites uh in server cube that form factor so it's 10 centimeters by 10 centimeters by 30 centimeters in CIA and not an actual satellite interviews but at the same chose the form factor and loading bring real satellites anymore and because and so our goal is to build satellites actually was sort of consumer-oriented their electron exerts an not space hardened electronics so the mean actually build them up quite inexpensively and but also being a small electronic so we have to build a small package instead of using in technology from the 19 eighties and nineties or something like that we're actually using current search technology and so we can build small and cheap and want a lot of so our steak isn't really to build the best satellites in the world it's the bill at the most efficient at sort of cost-effective satellites and fill certain roles the and so we talk about what the company we often refer to us as being a new space so the space this idea of commercializing space and doing things anyway instead of the old traditional way with like you 10 year long development programs and billions of dollars but it's also you can see things like space acts so there's a bunch of other worth cast and other folks that I would call a new space as well so basically trying to do space technology difference so and I come from this from other a traditional geospatial background and I'm not really know much of a space tackle 1 thing is really exciting for me working the labs is the fact that the changing space happened something that you know space aficionados in Germany have for a long time but everything I seem to go so slow so can remove government make it go faster but not the type of angry detail today we have launched approximately 7 the satellite so far which I'm a smaller proportion actually operational at this time and so that while impressive impressive but also the fact that we can have some satellites that have already yielded that of work is the thing that speaks to the fact that we can do it from small and cheap so we can afford it take risks and push fast things that I'm are also the goal is to be image in the world every day so by that I mean actually collecting 5 meter imagery of the world everywhere and the new fresh energy every day is of course a lot of but actually in the cloud cover stuff like that would mean the actual brand new daily mosaics that's my our hope is eventually rebuilding mosaic every day even if the parts under a cloud a actually energy from a few days before whenever the last part from Caffrey covers were operationally were quite a ways away from that and so would take hundreds of the Kunduz cell presumably to get that goal and we are doing things every day to improve our operational effectiveness non-smooth as a company are products are basically stand-alone images and these composite cylinder we talk about them composites mosaics with the sort of frequent timescale and ultimately derived information from that so things like change-detection another kind information products so ultimately it's too much data for people to be consuming directly get is will be boiling down into information products of that we would drill and look at particular things as appropriate the and I'm really they focused on mosaics students enough of a topic from the proponents of version that's about all you get to learn about the planet labs in general and 1 slide in
his picture of some of our down so the point well point of this slide is that we've actually gone through all went to different building designs over time and this is the every time every few months for launching satellites it's a slightly improved design from the last as opposed to design cycles in tradition aerospace might be in every 3 or 4 years and lodging the satellite with different designs what is so my
challenge and my part of the of the work I do that elapses try build the composite triangle build this global mosaic and so we obviously we would like it to be seamless and cloudless and radiometrically consistent so now we're building is of basically millions of scenes so the idea is that we have many many images of the same patch of ground and we can sort of pick through those either a pixel-by-pixel basis or a patch by patch basis to try and get the best pieces of each 1 were perhaps most current pieces of whatever is the appropriate criteria to build the school was it and so the scenes might be spread over weeks or months among these what to be of the economy dynamic decisions about how to build this mosaics of a submodular fire the thing would be a month was a taking the best images that have 3 months of the world but also available the non mosaic that might only be imagery from last week a few something like that and even if that means having to take lower quality was imagery but basically socially something more current in some cases we might be building products for someone ag sector where they really want to see just 1 piece of the growing season so we could say build them a mosaic of all of Brazil which is just the imagery from these 2 is 3 weeks and that tells you something very specific that picture of agriculture in the whole country I'm in a fairly narrow times so the the sort of uh in traditional Earth-observing you could do this with motors possibly Landsat over longer stretches but where I like to think they were the 1st people will be able to do it at this kind of frequency at around the 5 and concerns imagery also from different times of day potentially so we have a larger satellites into a variety of or it's were kind of the launches of opportunity in the sense that when we can get on our satellite or as a side cargo on as we'll take a wide variety of work it's the kind of ideal for us but that doesn't mean that the energy going into the mosaics and has a bright even conditions from different elevations means different resolutions also different in some of our satellite certain sink orbit which means they're always collecting roughly the same time of day for any given location on the Earth and other ones there are things like the space station orbit where you actually collecting at all times a day and from the city of funny an English using different lighting and stuff like that so what are the challenges is really to correct for that and obviously also from weather conditions
as soon is the part where I make fun of everyone else's music I have so we all underground in Google Earth there you know and also these other imaging was mosaics and and you get a little bit out of the city and suddenly the binding also courtly things 1 of the things actually motivated me you are when as considering the change the planet labs is I was can help a little bit with the OpenStreetMap mapping in the Philippines and there's actually some not bad imagery in being versus real on the coastline somebody obviously acquired was interesting parameter is going to be something we didn't to go far inland before here in the Landsat so uh these global mosaics at say the consumer level actually important for all kinds of processes for people understand about the places so not just for science not just a industry before people understand things and also to respond to disasters and so on and so forth so as an example right found very difficult to collect for about data in the land areas assist you cannot resolve things in sort of useful fashions sort also getting in final bit of other people's was X and also the bat mean backdrop we actually uses our base map or doing all our testing work uh the labs is the current that box satellite the fact that I have had some so and it's actually provide a lot of ways and they're really a guiding light to me that map outside satellite other ways as far as important techniques to apply modern some of the old stuff so because thinking clearly said 0 this just the old building there doing stuff good but so what was on make here is that this is an eerie Brazil and so it's not considered of high priority interests not United States so you still I claim that with imagery that's cloudy this is actually a part of Brazil that I've been using and a lot of practice on because the like it's almost always cloud it's almost always these popcorn clouds going over so if you take a strategy which is trying to find a cloud-free days time final cloud you seen is actually a very difficult place to do so during a large part of the year particular guess maybe it's the growing season so that they can the details and factors almost always popcorn class going worked out so so this is sort of typical that and I'm obviously there's been some effort that went into some point 3 collecting the best cloud-free emitted they had the time to build this was a good test of so that's really I think this is actually from lands that this later in the of this comes from being so this is actually
much higher resolution imagery essential and ways but clearly the points and is once again cloudy so this is actually a nontrivial nontrivial size town but it's near part of the world that people are not that interested in doesn't you know justify an aggressive acquisition costs and things like that the other thing that really what you see here is cloud shadows on the top of a little more but 1 of many years use of the the from in this case they're not really have the clouds to so literally really were that much with a cloud shadows but 1 of the real challenges putting together mosaic is isn't just getting rid of the classes getting rid of cloud shadows and the clouds even basically try not to get a light colored pixels prefer the darker pixels but if you take that to the extreme where C and similar examples you basically get a collection of projects so that is
also an example of a success so previously was working a Google and for the last year that's that has spent time on retention team and 1 of the things of the retention team does is global composite thing from Landsat 7 Landsat 5 data and really a lot of the techniques that I've been trying to apply a basically things I sort of remember from when I worked there another valid it is much smarter than me before can replicate some seconds but actually bring this up as a demonstration of the success warn you know is an extremely cloudy country so basically in this case they taken Landsat 7 data from about 12 years I think and to build this global what the color pretty pretty Earth image so it's basically Atlanta scale so it's 30 meter resolution but they taking basically 10 or 12 years of imagery and for every pixel that analyze that and they tried to pick the representative cloud-free pixel along with some advice other rules around trying to perform for a green time you're versus the every time it was actually successive this is near the world where if you were to look around you just would not find cloud-free images in fact you often if for typical inside acquisitions here almost all the images were value or at least and is always parts of it seem to be cloudy so for the lesson here is that you actually take a deep enough stack of good enough history enough imagery everywhere is uncovered at least now and then the the on the
other hand if you zoom in a Google Earth Google Maps into Borneo and you get down to the high resolution data using discover exactly the same old problem is that it's using more traditional dating techniques and wherever you go you start getting cloud that actually got a bunch of different fronts here they kind of patch together no the that you were in the not the screens and using kind of may goats and in a slightly clearer view that this is kind of have together from different images and you see similar clouds can cut off in the middle and if I see that you know you you have 1 image product in another yet a visible this is actually so I live in live
in a rural Canada and tell a few years ago so this is actually in the road that I lived on for narrowed road so there's always very disappointing for me to go into Google Earth Google Maps and in you know in a few hundred feet from my house there would be this nice for the data they must a graph in the county area the provinces somebody and then you know you go through a little strip of land that they then at my house in town and after I moved into town in the middle inset data for the time which is quite tragic so what you end up with is a mishmash of you know products coming from sorts of resources the basically the trying to use the best data they have habits they have some really beautiful photo data what you you've got all these arbitrary transitions and holes getting filled in with whatever resolution have so I completely understand how this happens inside given what they're trying to do this is in some ways the best they can do 1 of things I'm hoping is given a rich enough set of relatively consistent input data from our satellites we can that we we don't end up having to have this kind of patch effect especially about this you know varying resolutions problem although at the cost of have superhigh resolution considerable seems in different sources quite visible this is a another
reliance on direct products of the from the US GS partly I want to show here that even with Landsat data you can end up with the situations of visible what seen boundaries so basically the way the scenes are collected from a year the that and sort of acquisition boundaries it's also got terrible spreading effect but that's something to do with the sensor image and obviously very aggressively those holes or I guess what happened is that they've got that 1 scene with cloudy holes and then fill in the gaps from a scene that underneath see really you onstration effects in the again
so I now having made fun of them on suppose of this was to come up with a plan that's going to produce a better composite and my wish and when I see you know when I signed up for this presentation I thought I would be reporting on that is that I will be a recording and my attempts to do so and with you know limited degrees of success and lots of things yet to work on and I hope you will still be Molly interesting and get a sense of through an aspect of image processing that's a sort of still developing OK so and if I don't know all the talking mostly about their working the land data so violence has only like little 100 thousand seem so as we do not have a global coupling at Attica operation teams and particularly do have this deep stack ideas some ideas of the working the stacks of like 5 10 15 20 images deep so we certainly do not have that kind Catalonia so what I've been trying to do is experiment Landsat data basically developed techniques so that particular with a year a year a quarters worth of plants and it's so most of my examples all discusses that show a couple pictures of of our data at the end but that's not and we don't really have the data depth to do the full compositing approaches OK so the idea is whether the stack of imagery and I'm going to go through pixel-by-pixel so we carefully co-register Minkow registering is very important to me 1 of the whenever a processing chain problems and some and then so we go take 1 pixel at all the values that we have for that pixel from all the different scenes and then we can a certain process through them in various ways to try to get what we think is the most represented pixel in an optimal you go to the next pixel side and you do that complete analysis again independent of all the other so it's a highly separable thing you for every pixel you're actually trying to take the best thing that you could find in actually have you know a mishmash of clouds even with relatively small gaps in theory and you'd still be able to get some value to the data and again that's the sort of idea of patients the parameter when I saw the but different answer techniques and measures that we can take so In theory what we want is the ties qualities of various ways you can measure quality and if you are really just worried about avoiding clouds you can just take the debt dark aspects on basically that's gonna class avoid have clouds it also always he's so 1 of the problems we also have as different conditions so they can be high cirrus clouds everything and don't show up as like this puffy weight clear mass is just the general lightning in the scene that really clear boundaries and or you can have grandfather the sort of ground-level atmospheric effects and generally manifest as a lighter scene relative to other candidates from the same area so the sort the naive approaches most and talk about is taking the darkest pixel and then we obviously going to come back to the whole world Cup challenge issue and other categories and so for the the the Landsat prettier they did a google is actually a measure that was sort of a mixture of darkest and greenest there when you create collecting across the whole year or many years you want try and pick and you ideally want pick from an optimal seasons of green is generally the most interesting season but also it you don't want to make some green season and dry season all the things that we're interested in doing is on the verge of other to image quality issues some of which are a very across the sensor is some of which very scene by scene so when these is we have a very severe than getting effects and there a corner is scene image quality is very low so do I want to actually have a mask which says and I prefer to use the pixels from the center 1 scene rather than the pixels from the corner of another scene because we know that the centers there in the corner on you know if you have a little bit of us on the lands or other image anomalies and things like that you can apply the similar techniques so if you got and damage pixels etc. you can if you can have per sensor and quality characteristics you can use that to de-emphasize assigned you could maximum these runaway sometimes the best images you got for the spine is better than nothing but you definitely would prefer to take something that didn't have those problems and we also would like to build a mosque and clouds that were very confident of so this different techniques darkest will do a kind of automatically but if you're going for other techniques that maybe you want have explicit carbon tasks and get rid of the of and so that's also much of 1st seen characteristics so we would much prefer to have uh scenes that some well at time of day relative to close to the donor dust and even have much better signal-to-noise ratio and so even if things then you're done dust we correct for the sun angle in terms of his overall multiplier for the scene that the thing the noise is actually quite a bit worse a Sunday so these there's sort of scene level characteristics that sometimes we also have different senses which are are more or less and focuses about other things at that level wanna keep track of we prefer to take the best quality seems worry have them but we only have a lower quality scenes we would rather have mosaic then nothing will again some general criteria the preprocessing so we have a course the whole
processing chain the most important part of it is actually a rectification so I thought were taking these pixels and you might take were not doing the processing for our data and 4 . 7 year by 4 . 7 meter uh rectangles you basically you have absolute error relative spatial between your scenes of you know 5 meters and you basically not looking at the same spot on the Earth and all those pixels important evidence from registration so that's a big part of our preprocessing change won't going to here but as I had actually will we have a separate talk and have didn't get accepted as a whole again so on the case of lands that will basically take the you can imagine we just re-projected all to borrow put global grid which is located on the left side all processing plants that I'm at level 13 that's basically 90 meter pixels and luckily glance at a spatial accuracy is generally pretty good although I get a few scenes that some something increasing crazy and nobody noticed and really can't strange effects and don't course on having show that and when we do our own data I do it at 2 levels deeper so that the 4 . 7 so i have so also the processing chain that goes into that were a tribe you directly from was so that I can do just 1 resampling step to get rid of serving grid the reason I'm picking these murky things as we once a regular tiles ultimately so it's best to go from my last seen in 1 step to that the minimum damage the images so it also goes a variety of calibration steps which are working moderately well but not perfectly what is a serving and talk more
bustling city listening to give you the input areas of this is that that's a small a small patch of that era in Brazil initial before front-running give the sense that it is in fact most the image is quite cloudy so get 3 typical scenes since they've each got uh
you know popcorn castles at 3 typical sings the sum of scenes for this area that are basically all clouds this some that are more like you know me tended to decrease OK so that some typical
images going into and value can see quite
distinct cloud shadows that extend quite a ways away from the core and the technique is pixels stacking
see what they call so these are basically graphs of 1 pixel and in this case and graph against time and I got red green and blue so watch what I actually hope to see in these was something that was more of a seasonal in a compact inconvenience but he was gives the sense that here we basically got every 1 of these everyone is points is 1 the Landsat layers so you go a limited number of data points probably about at 20 most of which are off the graph which are basically super clouds lots of really bright it just goes off the graph aware that's not gonna be interesting but we've got still a fair number of data points so maybe 7 or 8 10 that seem to be useful here and I'm I don't think we have come back to me and looking at it this way is actually not all that helpful so you can't actually decide what is going be as quality loss quality very easily this so the next thing we do is actually sort and in this case and sorting by brightness so the brightest on the right which is you know off the graph is basically the cloud pixel again and then down here we've got various sort of usable pixel values no and when I when I should go back and I trace this carefully having to go through individual pixels and you can use these ranges of cloud the fringes of the cloud where his wispy you couldn't necessarily even detectors a class or to keep could conceivably even hazy although in this case this area wasn't prone to and wasn't that prone to like and he's so you probably is you get higher they're beginning to modern and not that I will proceed with 0 there's some sweet spot in the center which is uh well it's not in the shadow not not the easy pixels so that you the very most
naive thing I just take the darkest and so that for every
pixel i basically the sorting and just take the bottom and this is the kind of scene enough with the
moons in this break here the right but what you're really seeing here is all these things these in the scenario geometry these there this isn't a physical feature in the ground this is just cloud shadows of basically collected cloud shadows for my region of interest you get a few sort of well areas although tend to be on the downside as well that I think we can flash so this is from the base
layer and just to show that these this
captures the meaning of the grounds just a
cloud thing I see that you're introducing all wanted detailed me that's quite meaningless
um press that same graph again so I think turn the point now that we what we want to do then is that the in the dark and we have to was taken and so at the end of year instead of taking the darkest pixel or the right it's all we want to to take some sort of a median or percentile some way through the stack and hopefully were following someone into that's reasonable well at pixel at home and you can also think of this as a a as a liar removal the cloud to basically liars the collapse of the cloud shadows are basically outliers we want something is a represented the middle it also kind helps to avoid you know somewhere the process the other anomalies that could be something weird happened once to unseen and just by virtue of being sort of a central part % Alameda or something like that you're actually avoiding a lot of those strange results who
also I've experimented with a variety of different percentiles BC to collect so this an example being at the 85th percentile I will say someone actually 1st remove things that are known clouds the case of glance that it comes with a cloud mask which is we know this is cloud we think this might be clouded a chance is clouding were really sure this is not what I have done this before I do the percentile selection is approved the ones that are this is for sure class so the rest of it I don't want to remove this rather keep a richer dataset but it's at least can be the ones really classes 85 % after removing the ones that were already fully identified as cloud in the winter metadata and so on and this is a 70 per cent of a 50 50 so at the 85th percentile you can actually still see really strange blotchy cloud shadows and things like that are coming into from as you get down to 55 per cent of it's a little bit more normal but it's still really what is the image qualities in good here but in fact the underlying image is not that great so um as I go into it so that's the lesson is this is not that perfect technique and my
title before about the should be the sweet spot in the middle which is nice well at pixels when I was doing these graphs and doing is grabbing I was hoping it was gonna look like this so this is not real data this is my wish I was so I was hoping we could do this and I said no you know we are on the road on require 1 you know how and some cloud shadows and that was the Soviet thing I could be given enough data points I could actually go through and do this on a fairly regular basis there maybe 4 neighborhoods assuming that unfortunately it was not really born outside the scope wrote some tools so that I could pocket pixels in the facts and sea sea grass and stuff like that so here are some typical examples and so there isn't any real obvious breakpoints it's just now at accurately that little of 4 by 4 neighborhoods I get identifing saying well maybe I would see that I have more data points in these regions aside that been released basically not very much to to see here now part is also were taken Israel's as a different season so is a lot different you know effects of it going on so you don't it's not like we do that again
bring on a so nevertheless even though the technique was I the 65 percentile and I didn't generate an image of the world of Landsat and so this is the overview of that's a loan part of the challenge is just a readable of infrastructure and actually in the process of semantic data sources that zoom in
on Portland's from analysis you can really
start to see some problems using a zoom in on a farming area and I isn't look between ah mosaic
and the base map what you see is there's a lot of introduced basically garbage in the class and the fields a lot of detail is not real data from further might be mixing dry and and growing seasons on the part of it is just and the fact that this flipping back and forth
between scenes every pixel to get the stench things
again don't they would also find this concept the seasonality so I did a bunch of graphing of seasons so this is basically the number by this is necessary data over about a year and a half what I didn't see is a nice obvious number this likes it's relatively few data points of the patterns I might be you pick something and the and the analysis what is a percentile selection avoids a cousin needs and most Cochetel senescence goodbye you get we're mixing seasons and there's no really obvious breakpoint in which to do at the end of this terrible patchwork effect that what is it that
imagery and so this little that year is actually looking through the base map of the rest of it is I've seen so 1 point that making here is absolute calibration is important so we some of this light is over here is the use of mostly on word is still not getting all this seems to be quite consistent the basically collected in strips so the consistent within the strip club between strips were getting a greater
consistency and also show and also do things like a drop a where we know a cloud so we can and the edges it's hard to know the cloud but in the center we actually got that the most OK so
this is done with a thing called the pixel at some positive hazards to invigoration Pfizer g for reading writing data and it's got to defer to these quality measures and hopefully will be added cloud masking and we get also recorded which seen every pixel came from sort of source trees software open source and you have their as other easily it doesn't have a really good documentation I hope to some flush it with a bunch of examples using winter stuff in coming weeks and
you some examples of Jason file so this would give a list of positive so you can fool around with already 1 of the things in you give a different when we the word on the on the and this is the percentage of the time I have this is 1 from our own imagery so what is that the sum of these are sensor-based quality files so they get passed in his of that has some quality files covered dates and the basic and that's another simple example look at
the future directions from blending around the percentiles instead of picking 1 guy actually pick a few pixels close to that and blend them an actor Charlie late from that box of saying that we really have to glance at a whole improve the preprocessing obviously experiment with greens pixel from would soaring every pixel then completely independently and as much as I love that idea ideas a simplifying assumption and only it's gonna work so I'm from against our biasing to select from the same thing seems as your neighbors and tell a quality difference becomes 2 significant and if they're all fields and in the fall back to do using more traditional patch vector these defects and try and understand where the clouds are geometrically on the most interesting collaborating with others in Part of the reason open source this and I want have the stock was there are a bunch of people doing sort of composite in things like this in the world and I would love to get some sharing of experience and even code and I plan to raise more docks and I'm also hoping to make our land said the entitled form and that the mosaic of thank you very much
Ebene
Telekommunikation
Satellitensystem
Subtraktion
Bildgebendes Verfahren
t-Test
Versionsverwaltung
Zellularer Automat
EDV-Beratung
Information
Raum-Zeit
Computeranimation
Überlagerung <Mathematik>
Spezialrechner
Formfaktor
Prozessfähigkeit <Qualitätsmanagement>
Mosaicing <Bildverarbeitung>
Endogene Variable
Datentyp
Meter
Softwareentwickler
Optimierung
Stochastische Abhängigkeit
Bildgebendes Verfahren
Nichtlinearer Operator
Raum-Zeit
Bildauflösung
Strömungsrichtung
Biprodukt
Quick-Sort
Teilbarkeit
Auswahlverfahren
Satellitensystem
Energiedichte
Würfel
Mereologie
Prozessfähigkeit <Qualitätsmanagement>
Server
Information
Streuungsdiagramm
Kreiszylinder
Satellitensystem
Subtraktion
Punkt
Gerichteter Graph
Computeranimation
Eins
Demoszene <Programmierung>
Spezialrechner
Mosaicing <Bildverarbeitung>
Bildgebendes Verfahren
Bildauflösung
Diskretes System
Gebäude <Mathematik>
Orbit <Mathematik>
Ideal <Mathematik>
Biprodukt
Frequenz
Dreieck
Quick-Sort
Entscheidungstheorie
Arithmetisches Mittel
Rechenschieber
Energiedichte
Patch <Software>
Konditionszahl
Hydrostatischer Antrieb
Basisvektor
Dreiecksfreier Graph
Mereologie
URL
Varietät <Mathematik>
Satellitensystem
Bit
Punkt
Prozess <Physik>
Quader
Mathematisierung
Klasse <Mathematik>
Fastring
Übergang
Spezialrechner
Mosaicing <Bildverarbeitung>
Gruppe <Mathematik>
Abschattung
Vorlesung/Konferenz
Bildgebendes Verfahren
Bildauflösung
Softwaretest
Parametersystem
Pixel
Machsches Prinzip
Gebäude <Mathematik>
Güte der Anpassung
Ähnlichkeitsgeometrie
Teilbarkeit
Quick-Sort
Mapping <Computergraphik>
Arithmetisches Mittel
Flächeninhalt
Rechter Winkel
Mereologie
Strategisches Spiel
Projektive Ebene
Extreme programming
Streuungsdiagramm
Zentrische Streckung
Subtraktion
Sichtenkonzept
Pixel
Zwei
Selbstrepräsentation
Bildauflösung
Schlussregel
Google Maps
Biprodukt
Quick-Sort
Patch <Software>
Mereologie
Meter
Kantenfärbung
Bildgebendes Verfahren
Streuungsdiagramm
Touchscreen
Bildauflösung
Soundverarbeitung
Satellitensystem
Graph
Mosaicing <Bildverarbeitung>
Open Source
Gruppenoperation
Biprodukt
Ein-Ausgabe
Quick-Sort
Computeranimation
Demoszene <Programmierung>
Randwert
Patch <Software>
Programmfehler
Flächeninhalt
Menge
Direktes Produkt
Digitale Photographie
Ruhmasse
Bildgebendes Verfahren
Widerspruchsfreiheit
Bildauflösung
Bit
Prozess <Physik>
Extrempunkt
Extrempunkt
Computeranimation
Übergang
Spezialrechner
Meter
Einflussgröße
Nichtlinearer Operator
Parametersystem
Präprozessor
Kategorie <Mathematik>
Winkel
Ruhmasse
Zusammengesetzte Verteilung
Verkettung <Informatik>
Fehlerschranke
Registrierung <Bildverarbeitung>
Konditionszahl
TVD-Verfahren
Präprozessor
Charakteristisches Polynom
Pixel
Varietät <Mathematik>
Subtraktion
Gewicht <Mathematik>
Mathematisierung
Klasse <Mathematik>
Rechteck
Automatische Handlungsplanung
Geräusch
Kombinatorische Gruppentheorie
Term
Physikalische Theorie
Strömungsgleichrichter
Task
Demoszene <Programmierung>
Multiplikation
Weg <Topologie>
Datensatz
PERM <Computer>
Inverser Limes
Bildgebendes Verfahren
Analysis
Streuungsdiagramm
Soundverarbeitung
Pixel
Green-Funktion
Relativitätstheorie
Bildauflösung
Winkel
Bildanalyse
Quick-Sort
Verdeckungsrechnung
Programmfehler
Bildschirmmaske
Minimalgrad
Flächeninhalt
Parkettierung
Mereologie
Räumliche Anordnung
Streuungsdiagramm
Demoszene <Programmierung>
Patch <Software>
Gewichtete Summe
Flächeninhalt
Ein-Ausgabe
Streuungsdiagramm
Bildgebendes Verfahren
Computeranimation
Einfügungsdämpfung
Pixel
Punkt
Graph
Benutzerfreundlichkeit
Klasse <Mathematik>
Zahlenbereich
Ungerichteter Graph
Kontextbezogenes System
Quick-Sort
Spannweite <Stochastik>
Flächeninhalt
Rechter Winkel
Vererbungshierarchie
Abschattung
Inverser Limes
Speicherabzug
Bildgebendes Verfahren
Streuungsdiagramm
Demoszene <Programmierung>
Pixel
Flächeninhalt
Minimum
Kontrollstruktur
Abschattung
Räumliche Anordnung
Quick-Sort
Streuungsdiagramm
Computeranimation
Resultante
Zentralisator
Pixel
Punkt
Prozess <Physik>
Graph
Quick-Sort
Computeranimation
Programmfehler
Arithmetisches Mittel
Ausreißer <Statistik>
Mereologie
Abschattung
Streuungsdiagramm
Soundverarbeitung
Nachbarschaft <Mathematik>
Bit
Subtraktion
Pixel
Punkt
Klasse <Mathematik>
Ungerichteter Graph
Dialekt
Computeranimation
Verdeckungsrechnung
Eins
Metadaten
Trennschärfe <Statistik>
Mereologie
Basisvektor
Abschattung
GRASS <Programm>
Streuungsdiagramm
Bildgebendes Verfahren
Varietät <Mathematik>
Prozess <Physik>
Flächeninhalt
Mosaicing <Bildverarbeitung>
Open Source
Mereologie
Zoom
Bildgebendes Verfahren
Computeranimation
Analysis
Formale Semantik
Streuungsdiagramm
Soundverarbeitung
Punkt
Pixel
Klasse <Mathematik>
Zahlenbereich
Zusammengesetzte Verteilung
Ungerichteter Graph
Mapping <Computergraphik>
Demoszene <Programmierung>
Graph
Datenfeld
Trennschärfe <Statistik>
Mustersprache
Mereologie
Mixed Reality
Analysis
Mapping <Computergraphik>
Punkt
Vorlesung/Konferenz
Betrag <Mathematik>
Wort <Informatik>
Tropfen
Streuungsdiagramm
Widerspruchsfreiheit
Streuungsdiagramm
Gewichtete Summe
Pixel
Ortsoperator
Open Source
Hasard <Digitaltechnik>
Quellcode
Elektronische Publikation
Quick-Sort
Computeranimation
Verdeckungsrechnung
Verdeckungsrechnung
Spezialrechner
Open Source
Einheit <Mathematik>
Software
Wort <Informatik>
Pixel
Streuungsdiagramm
Einflussgröße
Subtraktion
Präprozessor
Pixel
Quader
Green-Funktion
Gemeinsamer Speicher
Open Source
Patch <Software>
Vektorraum
Marketinginformationssystem
E-Mail
Quick-Sort
Code
Computeranimation
Richtung
Patch <Software>
Mereologie
Richtung
Pixel
Streuungsdiagramm
Vorlesung/Konferenz

Metadaten

Formale Metadaten

Titel Compositing a Global Mosaic
Serientitel FOSS4G 2014 Portland
Autor Warmerdam, Frank
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 10.5446/31705
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 Planet Labs is collecting images from dozens of satellites in order to build timely global cloud free mosaics at around 5 meter resolution. I will review the software components we use to accomplish this, as well as discussing challenges and solutions in this process. With luck I will be in a position to show off our global mosaics, and offer Planet Labs open source compositor software.
Schlagwörter cubesat
mosaic
global

Ähnliche Filme

Loading...
Feedback