We're sorry but this page doesn't work properly without JavaScript enabled. Please enable it to continue.
Feedback

Automating Image-Based Boundary Delineation

00:00

Formal Metadata

Title
Automating Image-Based Boundary Delineation
Subtitle
The BoundaryDelineation QGIS plugin
Title of Series
Number of Parts
295
Author
Contributors
License
CC Attribution 3.0 Germany:
You are free to use, adapt and copy, distribute and transmit the work or content in adapted or unchanged form for any legal purpose as long as the work is attributed to the author in the manner specified by the author or licensor.
Identifiers
Publisher
Release Date
Language

Content Metadata

Subject Area
Genre
Abstract
As intention to facilitate faster establishment of cadastral databases, different projects were initiated to support rapid automated approach for boundary extraction from remotely sensed imagery. Such project is the EU funded “its4land”, developed as a joint venture between several organizations including University of Twente. Despite the current advancement of the modern segmentation algorithms, there are imperfections in the output that have to be further manually filtered out. Therefore, a QGIS plugin called BoundaryDelineation was designed and developed to support an interactive delineation of visible (cadastral) boundaries to speed-up the process. The plugin supports the full pipeline of loading, picking and drawing segments, and saving to a remote server the final validated output, which allows rapid reduction of the mouse clicks compared to manual delineation, which affects the overall speed of creation of cadastral entities. The presentation will consist of a general overview of the its4land project, and main focus on the usage of the BoundaryDelineation QGIS plugin as a more general-purpose assistant in boundary delineation.
Keywords
Computer-generated imageryPlug-in (computing)Computer fontThermodynamisches SystemBuildingThermodynamisches SystemComputer fontStudent's t-testXMLComputer animationLecture/Conference
Plug-in (computing)Thermodynamisches SystemComputer-generated imageryComputer fontPresentation of a groupArchaeological field surveyComputer programmingInformationType theoryMetreMedical imagingSelf-organizationMultiplication signSoftware engineeringXMLComputer animationLecture/Conference
SoftwareRaster graphicsWeb browserProjective planePlug-in (computing)Flow separationThermodynamisches System2 (number)Set (mathematics)MereologySoftware engineeringSatelliteData managementMultiplication signComputer fontWordMappingContext awarenessMedical imagingProduct (business)Open sourceTheory of relativityReliefNetwork topology1 (number)Computer animation
Shared memoryVisualization (computer graphics)AuthorizationThermodynamisches SystemComputing platformMereologyMedical imagingData storage deviceNetwork topologyArithmetic meanSharewareUniverse (mathematics)Projective planeComputer animationLecture/Conference
Data typeComputer programInformation and communications technologyBuildingThermodynamisches SystemImage resolutionComputer fontProjective planeHorizonPoint (geometry)Extension (kinesiology)Line (geometry)Image resolutionMedical imagingThermodynamisches SystemNeuroinformatikMereologyAlgorithmProcess (computing)Computer fontLocal ringTerm (mathematics)Canonical ensembleComputer animation
Thermodynamisches SystemDiscrete element methodComputer fontLine (geometry)Computer-generated imageryInteractive televisionPredictionRaster graphicsPlug-in (computing)AreaOrder (biology)Thermodynamisches SystemEndliche ModelltheorieService (economics)Medical imagingForestOperator (mathematics)Reading (process)MereologyModule (mathematics)DigitizingRandomizationPlug-in (computing)OrthogonalityPanel painting
Asynchronous Transfer ModeUniform resource locatorObservational studyComputing platformFluidComputer fontGUI widgetRaster graphicsFunction (mathematics)Interface (computing)Plug-in (computing)Plug-in (computing)Raster graphicsLine (geometry)Thermodynamisches SystemInterface (computing)CAPTCHAOperator (mathematics)Medical imagingContext awarenessProcess (computing)Lie groupProjective planeOpen setXMLComputer animation
EmulationWeb pageDot productDirection (geometry)AlgorithmLine (geometry)Vertex (graph theory)outputExterior algebraLie groupComputer animation
Computer fontGUI widgetPolygonLine (geometry)Plug-in (computing)Interface (computing)Structural loadInformation managementDirected setDisintegrationComputing platformProgrammable read-only memoryModal logicProjective planeCategory of beingComputing platformRule of inferenceRow (database)Power (physics)INTEGRALLie groupResultantMenu (computing)Set (mathematics)Line (geometry)PolygonComputer animation
Plug-in (computing)Polygon9 (number)Hydraulic jumpVideoconferencingPolygonSquare numberoutputTouchscreenLine (geometry)Computer animation
Menu (computing)Hill differential equationGraphic designVertex (graph theory)Insertion lossLine (geometry)Square numberPolygonShape (magazine)AlgorithmVertex (graph theory)Computer animation
TopologyMereologyVertex (graph theory)AlgorithmPerfect groupSquare numberMaxima and minimaRectangleWeightThermodynamisches SystemLengthPolygonMultiplication signRight angleLine (geometry)Condition numberComputer animation
Computer wormMIDIMenu (computing)DisintegrationComputing platformDefault (computer science)Field (computer science)TopologySlide ruleComputer fontAlgorithmVertex (graph theory)Line (geometry)PolygonOrthogonalityRight angle2 (number)Lattice (order)Field (computer science)Network topologyProjective planeComputing platformString (computer science)Functional (mathematics)Combinational logicINTEGRALThermodynamisches SystemDot productPatch (Unix)Data storage deviceArmWordProgrammable read-only memoryAdditionFreewareAddressing modeAnalogyAddress spaceProfil (magazine)Computer animation
Band matrixMedical imagingLink (knot theory)Mathematical optimizationThermodynamisches SystemSpectrum (functional analysis)Matter waveGraph coloringShift operatorDifferent (Kate Ryan album)Image resolutionAlgorithmOrder (biology)Execution unitSinc functionBitAreaPixelResultantMathematical analysisService (economics)MetreFood energyArmCircleCloningMultiplication signCategory of beingEndliche ModelltheorieLecture/Conference
PredictionLine (geometry)Computer fontThermodynamisches SystemInteractive televisionComputer-generated imageryUniform resource locatorObservational studyComputing platformEmailDemonAreaSummierbarkeitTopologySuite (music)Projective planeAdditionAreaPixelSatellitePlug-in (computing)AlgorithmMultiplication signState of matterPanel paintingXMLComputer animation
Slide ruleComputer fontMultiplication signPresentation of a groupComputer animationLecture/Conference
Transcript: English(auto-generated)
OK, we'll start with the next presentation, which will be delivered by Ivan Ivanov. I heard it's a common name in Bulgaria. And he's a student at ITC, working on his masters.
And he's going to present on boundary delineation plug-in for QEs, where you can extract information from aerial images. Please. Hello, everyone. Thank you for coming. Thanks to the organizers for their studentship program.
My name is Ivan Ivanov. I'm from Bulgaria. And let's see what is the title of my presentation. A quick introduction, who am I? I'm doing my second year masters in ITC in spatial engineering in the Netherlands. I already have six years of experience as a software engineer using plural and JavaScript mainly
and recently Python. And yeah, I'm 25 years with the spirit of geographer. In my spare time, I have several projects, side projects related either to geo or to education. I have three. I'm contributing to three plug-ins. The first one is for handling the raster non-data.
The second one is for browsing satellite imagery. If you are not familiar with the protocol of stack, I highly recommend it. It's really convenient. It's something I was needing for a long time. And the boundary delineation plug-in that I'm going to present in a few slides. So the project I joined at ITC, it's called Eats for Land.
I joined recently six months ago. The project itself aims to deliver a set of land tenure management tools that should be cheap, fast, and easy to use. All these marketing words is there.
The idea is to be open source. And also the target place where it should be used is mostly in developing countries. We have partnerships with Rwanda, Kenya, and Ethiopia. And it's a big consortium around it.
So the main technical part is in the middle of the picture on the side. So first, people are creating sketch maps because they have the local tacit knowledge where is whose land and how big it should be and all this local knowledge.
Then it's uploaded and processed and digitized and used later to reference the extracted boundaries. The second part is flying create, which is you fly the drone, take the images, then create orthophoto. And stop here and automate it if you need it.
You can take this orthophoto and the DSM and extract boundaries, visual boundaries from the image. And finally, the platform publishing share where you store the data and the local authorities can access it.
The things that are surrounding in red are actually parts where IDC is involved and I'm also involved. As I said, it's an international consortium. There are three universities, the University of Fente, Munster, and Leuven. Hanseluth built is building the bigger part
of the technical part of the project and also local partners in Africa. The project is funded by Horizon 2020, by European Commission. So, what we are trying to solve. Because these countries are to some extent lacking behind
and they don't have established cadastral system. They usually use the traditional approach of taking a high resolution image and manually delineating point by point, which can be very slow process and sometimes frustrating.
Because sometimes you see an obvious line and you say, okay, can computer or algorithm help us with suggesting, okay, there is a boundary. So, this is what we are doing. We're flying the drone, getting the orthophoto, reading the algorithm and extracting the boundaries.
So, basically this is how it works. The ortho image from the previous module in the whole system. We have RGB image and the digital service model of the area of interest. Then we run the image segmentation using MCG.
It proposes the segments. Then there is a random forest, classificator to see which is more probable to be a boundary or less. And then it comes my part of the QGIS plug-in that operators are finally confirming,
okay, this is a boundary and it goes to the system. So, yeah, in the end you have cadastral boundaries. What is the It's Proland platform? It's important to mention here because it's kind of the central place of the It's Proland project, is the place where the data is stored and managed.
All these tools I told you, for example, for creating orthophoto, it's a dockerized open drone map. The tool that is extracting the boundaries also should be a dockerized image.
However, it uses some MATLAB algorithms, which makes it not very easy to do so. But we are looking for a solution in the future. So, the plug-in itself, it's very simple, simple interface. It's written in such a way that it's not mandatory
to be used only in It's Proland context. On the second line you select the segments layer, which is basically a lines layer that is the possible boundaries. You can also select a raster based layer for visual confirmation of the operator. You click process and you get interface
something like this. I'm not sure how visible it is. But the green lines are what's the algorithm proposed or the other input somehow. It can come not only from our algorithm that we developed but also from alternative. For example, my masters is also going to this direction.
So, probably next year I can also create such segments layer. There are little red dots that are the vertices where two or more lines are meeting. I'm not sure if it's visible here. So, then we go to step two, where you have four different ways to select the polygons of the properties.
Polygons, lines, vertices, and menu. I'm going to show them one by one. Just to mention, there is native integration with It's for Land platform. So, the projects you have on the platform,
you can load them directly from QGIS. Also, the segmentation lines can be listed there. However, this set does not have, there is no sets found for this project. And also, you can directly upload the final result to the It's for Land platform.
Okay, unfortunately, the video is not working here, but I can move to the next slide and show you how to select by polygon. So, unfortunately, it's not very easy to see, but there is a little orange square in the middle of the screen
that selects several obvious polygons. However, the layer we have as input is line layer. So, it just looks for surrounding polygon, and you have the surrounding polygon. When you are ready, you click accept down on the left.
The other possibility is to select by line. So, you have these different lines that are visually connected, but as a feature, they are actually separate. So, you can select two lines. There is also one orange square,
I'm not sure if it's visible. So, we're selecting these two lines, and it finds the shortest path how to close these lines. It's important to know that if you have more than two gaps, of course, the algorithm says, okay, I cannot find the best shape because it does not know how to close it.
But you can see in the bottom, the line is straight, not following the line that is suggested by the segmentation algorithm. Also, there is possibility to select by vertices. So, you can select vertices that should enclose a polygon.
For example, here, again, there is a little orange square or a rectangle that selects some of the vertices. And it finds the shortest path. Or, if you look at the plug-in interface,
there is one drop-down where it writes a boundary. Here, you can select different weights for the algorithm to make the shortest path. It can be either length or some weight that is produced by the algorithm that produces the suggested boundaries, or whatever else.
And the final part is, of course, the algorithm is not perfect. It reaches something like 80% accuracy in perfect conditions, it's the maximum. So, sometimes you have to manually further edit the lines. So, you select the polygon,
then you click on edit, there is one, edit, and you can edit the vertices that are inside this polygon. So, now they are selected, I'm not sure if you can see the blue dots. We remove them, and let's say we're happy with this polygon. Okay, this is the property that we're going to put in the Cadastro system.
Okay, we accept it, but then, if we don't update the topology, the problem would be that the little patch that is in the middle is not going to be selectable. So, if you click update edits, you can create another polygon around it,
and then you select only this patch that was left. And, of course, you can select further polygons to merge them and to create a final feature. When you're ready, of course, you can click on finish. It's important to know that the It's for Land platform
by itself stores the features as line strings. For some reason, I'm still not very sure why Hasselt build decided to do it like this, but you can always click on a polygonized layer and use the native QGIS algorithms
to create polygons from the lines. So, basically, this is the functionality. It's not much, but it's a combination of already existing QGIS functionalities that are made easy to use. What is coming? Better integration with It's for Land platform.
Recently, there were new APIs that were published from our colleagues in Munster. So, loading base layer now wouldn't be the big ortho photo, but it would be tout layer. Also, saving features has improvements in the new API.
I have to write it in the next month, I guess. Currently, there are no fields that you can enter for the final layer. This is something I'm planning to add. For example, name, address, or whatever, or some ID.
And the topology problem I have is the algorithm that I wrote is not the fastest right now. It takes a second and a half or two seconds to update. I already have an idea how to optimize it, but I have to do it. Thank you for your attention.
If you have questions, you're welcome. Something I forgot to mention, special credit to Sophie Kromelink. She's the PhD who wrote the algorithm for extracting the boundaries, and actually, that's why I'm involved with this project.
She also suggested me to come here to present this, and yeah, I hope it was interesting for you. Yeah, we've had Sophie presenting her algorithm at other meetings already in Denmark, I think. Any questions on the work we do?
It's more on the algorithm side. How do you decide what algorithm to use for detecting the boundaries? Do you have some knowledge about it?
Actually, this is the research of Sophie to select the most optimal algorithm for extracting the boundaries. If you look at her GitHub history, actually, the third link there, she tried several different, and I think she finally settled with MCG.
Other questions?
We have plenty of time, yeah. So just like a general question, why would you need such high resolution for this,
since parcels are generally quite large, I assume, like I don't know what is the smallest unit that you have, and you need to map such large areas. It just would make sense to use some, for example, the three-meter planet imagery, or something that is a bit more lower in resolution. It doesn't seem like you would need to fly a drone over the whole country, just as a general question.
Yeah, actually, the data that you see here is with five centimeters resolution. It is proved to be with the best results. Unfortunately, the whole tool depends on visual boundaries.
So if you have very large pixels, and you have, for example, just, I would say, fence between two properties, then this is not going to be visible in the very large pixels.
Questions? What about using multiple bandwidths,
like not only sRGB bandwidths, when you do the analysis on a war resolution images? Sorry, can you repeat the question? It was related to what he asked about using the three-meter image resolution. What if you use the other bandwidths that are in the image in order to do the boundary detection?
Because I assume the algorithm that you're using is most likely only the visible bandwidth, that's the sRGB. So what if you utilize the bandwidth that the other wavelengths that are recorded in the image? Yes, to be honest, I really don't know why we consider
only the three visible spectral wavelengths, but I'm not sure if I mentioned it. We also use the digital surface model to, so it's not only seeing the shift between the colors or whatever,
we're also looking at the difference in the DSM to detect the boundary. But why we don't use, for example, let's say, microwave, I don't know. I really don't know, cannot answer. In addition to the previous question,
this algorithm and this plugin also can be used in urbanized areas. And especially in these three countries where the project is, where our partners are based, there are a lot of slum areas where the housing itself sometimes is,
the house is less than one pixel in, let's say, Sentinel or whatever else, satellite imagery. Do you have any final questions?
Even more, we have plenty of time. If not, then thanks again, and we have a lot of time to switch to the next one, 12 minutes. There will be another presentation on processing on QPs.
Thank you.