inous - Building an Eco-System for Indoor Spatial Information
This is a modal window.
The media could not be loaded, either because the server or network failed or because the format is not supported.
Formal Metadata
| Title |
| |
| Alternative Title |
| |
| 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 | 10.5446/43460 (DOI) | |
| Publisher | ||
| Release Date | ||
| Language |
Content Metadata
| Subject Area | ||
| Genre | ||
| Abstract |
| |
| Keywords |
FOSS4G Bucharest 2019124 / 295
15
20
28
32
37
38
39
40
41
42
43
44
46
48
52
54
57
69
72
75
83
85
87
88
101
103
105
106
108
111
114
119
122
123
126
129
130
131
132
137
139
140
141
142
143
144
147
148
149
155
157
159
163
166
170
171
179
189
191
192
193
194
195
196
197
202
207
212
213
214
215
216
231
235
251
252
263
287
00:00
InformationBuildingSpeciesArithmetic meanLecture/Conference
00:23
Physical systemInformationProjective planePresentation of a groupFlow separationComputer animation
00:44
Data managementConstructor (object-oriented programming)Process (computing)Standard deviationPositional notationProjective planeStability theorySpacetimeWeb serviceShared memoryProduct (business)Level (video gaming)Presentation of a groupUtility softwareMilitary baseMultiplication signCartesian coordinate systemCASE <Informatik>Computer animation
02:40
Computing platformInformationElectric currentStandard deviationPhysical systemData managementMobile appLevel (video gaming)DatabasePosition operatorCartesian coordinate systemShared memorySpacetimeSign (mathematics)QuicksortWebsiteWeb serviceState observerProjective planeXMLProgram flowchart
03:36
Physical systemStandard deviationWeb serviceEstimationTopologyData managementDisintegrationProgrammable read-only memoryDependent and independent variablesInflection pointAttribute grammarCircleVideo gameCovering spacePoint (geometry)Point cloudCycle (graph theory)Local GroupVector spaceModel theoryTrianglePolygon meshAuthoring systemGroup actionProjective planeConstructor (object-oriented programming)Level (video gaming)Point (geometry)RobotRule of inferenceComputer animation
04:19
Authoring systemTopologyDisintegrationData managementStandard deviationProgrammable read-only memoryDependent and independent variablesInflection pointAttribute grammarVideo gameCovering spacePoint cloudPoint (geometry)Web serviceWorkstation <Musikinstrument>Vector spaceTrianglePolygon meshCircleModel theoryLocal GroupGroup actionProjective planeLevel (video gaming)Point cloudPlanningProcess (computing)Lie group2 (number)Cartesian coordinate systemHypermediaWorkstation <Musikinstrument>Standard deviationSet (mathematics)Computer animation
05:17
Vector spaceSimultaneous localization and mappingTopologyLocal GroupModel theoryTrianglePolygon meshSoftware testingDependent and independent variablesWeb serviceInformationEndliche ModelltheorieDependent and independent variablesMedical imagingLevel (video gaming)Semantics (computer science)Cartesian coordinate systemProjective planeOpen setGroup actionMetropolitan area networkPlanningCausalityProduct (business)Web serviceComputer animationProgram flowchart
06:27
Software testingDependent and independent variablesWeb serviceGeometryTopologyAuthoring systemSystem callData managementCellular automatonAttribute grammarLocal GroupPoint (geometry)Simultaneous localization and mappingLevel (video gaming)Cartesian coordinate systemSimultaneous localization and mappingOpen setProjective planePoint cloudStreaming mediaRevision controlDataflowProduct (business)PlanningRight angleWeb serviceMetropolitan area networkPoint (geometry)Computer animation
07:25
Local GroupMenu (computing)GeometryDataflowPlanningoutputMedical imagingRevision controlGroup actionText editorLevel (video gaming)Engineering drawing
07:54
Local GroupOpen setLevel (video gaming)Noise (electronics)Streaming mediaPhysical systemJava appletMedical imagingMultiplication signPoint cloudInformation2 (number)Presentation of a groupRight angleArchitectureConnectivity (graph theory)Engineering drawingProgram flowchart
08:49
Polygon meshLocal GroupPoint cloudPoint (geometry)ArchitectureSpacetimeComponent-based software engineeringBeer steinOutlierNoiseGeometryTexture mappingSpacetimeSimultaneous localization and mappingCategory of beingType theoryPoint cloudArchitectureProcess (computing)GeometryNoise (electronics)Flow separationReflexive spaceMedical imagingPoint (geometry)Social classTexture mappingException handlingPresentation of a groupGame theoryRoboticsConnectivity (graph theory)Beat (acoustics)
10:24
AdditionLocal GroupPoint cloudPoint (geometry)GeometrySurfacePhase transitionProcess (computing)Computer animation
10:52
Local GroupText editorGeometryModule (mathematics)Process (computing)Data managementModel theoryAsynchronous Transfer ModeDatabaseInflection pointAttribute grammarExtension (kinesiology)Computer networkSpacetimeTask (computing)FreewareLecture/Conference
11:08
Local GroupText editorResultantMultiplication signComputer animation
11:28
Local GroupText editorGroup actionRight angleAdditionComputer animationLecture/Conference
11:44
Module (mathematics)Extension (kinesiology)GeometryAttribute grammarTopologyLocal GroupProcess (computing)Data managementModel theoryInflection pointDatabaseData conversionJava appletImage resolutionText editorControl flowClient (computing)Computer networkEstimationCellular automatonDependent and independent variablesCountingWindowInformationSpacetimeGroup actionLevel (video gaming)Cartesian coordinate systemDependent and independent variablesInformationData modelRoutingAdditionSemantics (computer science)File formatNeuroinformatikNetwork topologyMultiplication signInflection pointPoint (geometry)2 (number)Metropolitan area networkSoftware developerProgram flowchart
13:35
Local GroupDependent and independent variablesThermal expansionSimulationCartesian coordinate systemNumberCASE <Informatik>BlogMultiplication signInternetworkingInformationDynamical systemEndliche ModelltheorieEngineering drawing
14:55
Local GroupSmartphoneInterface (computing)User interfaceAndroid (robot)Cartesian coordinate systemPersonal digital assistantLevel (video gaming)outputProjective planePhysical systemMetropolitan area network
15:43
Open sourceOpen setPhysical systemSpacetimeWeb serviceComputer networkSoftwareProjective planeRight angleLimit (category theory)Open sourceCASE <Informatik>Multiplication signLine (geometry)XML
16:49
Dependent and independent variablesLocal GroupSpacetimeCASE <Informatik>MassPosition operatorComputer configurationMusical ensembleEvent horizonArithmetic progressionEngineering drawing
17:51
Electric power transmissionCASE <Informatik>QuicksortCountingField (computer science)NumberUniform resource locatorAreaMassCodeType theorySingle-precision floating-point formatLecture/Conference
Transcript: English(auto-generated)
00:08
OK, welcome everyone to the second talk of the session. We have Kijun Lee talking about building an ecosystem for indoor spatial information. Thank you.
00:20
All right, so this is quite a long story, but I'm not quite sure that I can make it within 20 minutes. But anyway, I will try that. All right, so this is a presentation about a project called INOS. INOS means indoor, I forgot it.
00:41
I name it, but I forgot it. Anyway, so we launched that project two years ago with several other teams. And starting from the analysis, investigation of why indoor spatial services, what technology, is not yet booming up.
01:01
So I have been working for indoor space more than 10 years ago, like 2007. At the time, I believe that indoor spatial services technology will be booming up very quickly within two years. But after 10 years, almost nothing happened. So why? What is the reason?
01:20
So we started the project by investigating the reason why there is a bottleneck of indoor spatial technology. So one of the reasons is there is no stable indoor positioning system. So GPS is out there, but not working in indoor space.
01:40
And the second reason is the cost for building indoor map is quite expensive, as they explained. The first presentation gave a very nice work about making the indoor map. But anyway, the cost is very expensive compared with outdoor space. And the next one is for outdoor space,
02:02
we have quite a very nice standard and hold the process. Sorry, I have, OK, I have to, OK, I have stay here. All right, OK, thank you. Anyway, so for outdoor space, there
02:22
are some ecosystem around standard. But for indoor space, from the construction of data to the managing and sharing and utilization application, there is no standard process for data production and the sharing.
02:41
And the final reason is there is no killer app. So the only thing that we are talking about indoor space is indoor navigation. But this is a very, very tiny portion of indoor spatial technology. So we have to find some solution for each problem, except the first one.
03:00
Because indoor positioning is, I'm working for the database. I'm the database guy. But indoor positioning has nothing to do with the database. So I decide to forget about indoor positioning. I try to use the existing indoor positioning method, but try to solve the four solutions.
03:22
OK, so based on this observation, we decide to set up an ecosystem from building an indoor map and managing it, sharing it, and developing some application for indoor space, more than a simple indoor navigation. So we started the project two years ago
03:42
and named it as INOS, INOS Indoor Outdoor Spatial Data Services. And occasionally, so INOS means it is also a name of a butterfly only in Australia. So we decide the logo like that way. So I paid a lot for designing this logo.
04:03
Anyway, so this is the goal and the scope of the project, starting from the indoor map construction. So there are many teams working for that. So this is group one. And group two is after we have the indoor map,
04:27
then we manage and process it to complement the data. And then after that, once we have the full data set of indoor space, then we can make a useful application.
04:41
So these are the main goal and scope of the project. And this is group one, group two, and group three. And we decide to set up an ecosystem of indoor space, indoor map in indoor GML, which is OGC standard. Actually, I am working as a chair of indoor GML in OGC.
05:03
So this explains the whole process of our project. Group one, we collect the data from the point cloud like stationary LIDAR or back pair LIDAR or even a tangle, like the very personalized device. This is one pillar.
05:21
And the second approach is from the simple floor plan, of image floor plan. But this is quite useful. So when I talked with some guys in Hong Kong or Singapore, because they have a lot of old buildings where they don't have any CAD data. So the only data that they have is floor plan.
05:40
So we have to sometimes start from the floor plan. So we vectorize using deep learning. And the extrusion to 3D, then we get 3D solid data. Then we convert it to semantic model. Because indoor GML, it is not only geometric model, it is also semantic model. So we include a lot of semantic information to this data.
06:04
And then finally, we get indoor GML. Then after that, once we have indoor GML data, we may apply it to several applications. But because this is the project funded by government, so we try to find some solution for public.
06:21
The first one is, of course, indoor safety, like indoor emergency response. And the second one is a voice indoor map for visually impaired person. And we try to make a service like open stream map for indoor version. We call it Open Indoor Map. So these are three main application
06:41
that we are developing as a group three. All right, this explain the indoor map production. But I will skip it. I will explain it. Anyway, so these are the left side is the methodology. And the right one is the approach that we are working for this project.
07:01
So first one is we try to make indoor map data from the point cloud collected by SLAM. And like the open stream map or the flow plan map, we can get indoor data. All right, so I would like to explain it from the,
07:23
so this is another, okay. So this is a team one of group one. That means once we have the input of a flow plan image map, then we try to build the geometry from that one is a 2D. Then we extrude to 3D and if possible,
07:42
we need some kind of a manual work. So this is something and I'd like to show that we develop, where is my mouse? All right, okay. I'm sorry that it is a Korean version of JOSEM. JOSEM is a open stream map editor, Java based open stream map data. Anyway, we loaded the image map like that.
08:08
Then we, so for the time being, we don't have any very nice tool to remove the some information, which is not really needed by our system. But so anyway, we remove some noise data,
08:23
then we try to build by using deep learning. And then we can make it and we match with the indoor open indoor open stream map. All right, then we do some other additional edit. And finally we make, okay, anyway,
08:47
so by doing so we can make indoor JML data. And the second approach is point cloud as the previous presentation. So we collected the data from point cloud, but what we are doing, what we do,
09:01
what we have done is we classify the data into two category. First one is architecture component, that means the wall and ceiling and floor. And the other class is non-architectural component, chairs and so on. So we can do that by using several types of method.
09:20
And then we convert it to indoor JML for non-nevigable space or nevigable space. These are the indoor JML terminology. So these are the process. So I will show the whole process, okay.
09:42
So from the SLAM, SLAM robot. So these are the SLAM that we develop by one team of our research initiative. And then we get the point cloud data. So it is quite a complicated point cloud data. Then we remove the noise data. We have a lot of noise because of the transparent glass
10:04
or reflexive floor and so on. So it is inevitable. So we collect the data and we apply almost the same methodology, but except PointNet. We conclude that PointNet is not that good, sorry. So we apply RANSAC and other type of deep learning. So finally we get the geometry
10:22
and we do some image data, image texture data. Then even we have the geometry, but we found that there are some missing things. So we have to edit it in manual way. So these are process of manual way. And sometimes when the door is blocked, door is locked, then we cannot enter into the room.
10:42
So we have to draw by, okay, where is the... All right, so I can, sorry. So, okay, so I'm sorry that there is some problem.
11:11
All right, then by doing by this one, we can show the final result.
11:25
Okay, I will skip it because it is, it takes time. We don't have enough. Oh, where is my slide?
11:41
All right, we get it. Okay, then the group two, we process the data that we get from the group one. And in addition to the data from group one, we convert the data from a BIM model, IFC data and CAD data because IFC contains a lot of semantic information.
12:01
So we convert it to indoor GML. And then we define the data model and we edit, we include some of the meaningful data like topology data or POI or sensor data. Then finally, we can get a very nice indoor GML data.
12:21
And then after that, we utilize the data for the application. One is open indoor map. So open indoor map is a application that develop on, it's a kind of geo-portal developed on the top of cesium. So we convert it to the data to F4D
12:43
but that GLTF of native cesium format because GLTF is very nice for outdoor but it takes a lot of time for visualizing for indoor space. Anyway, so we develop some other format. So we display it using indoor map
13:02
and then we can also for the application, sometimes we need a route computation not only for the empty space but also when we have something like an obstacle here, we have to consider this one, particularly for fire brigade,
13:21
they have to consider the obstacle. So we try to find the optimal route considering this obstacle as well. And the next application is fire response but I'd like to show, I'm sorry,
13:41
how many minutes I have? Okay, so this is a very interesting application of the indoor GML that we have. We have that indoor GML data. Then using a tool called the FDS, fire dynamic simulation which is developed by NIST,
14:01
we can simulate the expansion of fire and smoke. And the red dot, it means the persons. So persons try to escape in case of fire. So there are many model of the movement of the crowd. So initially before the case of fire,
14:21
we count the number of person of each room but after fire, because of the block of the internet and so on, we don't know exactly. So we have to make some simulation to estimate the number of person of each time. So by doing so, the fire brigade,
14:41
the firefighter can estimate how many person are there for each room. So this is very important information for firefighter. But unfortunately, these are some casualty. And the next one is, as I said, this is the indoor voice map for visually impaired person.
15:05
This is a smartphone based indoor voice map. If you have a smartphone like Android or iOS, they provide called the voice assistant user interface. So we call voiceover for iOS and TalkBack for Android.
15:24
But normally, the visually impaired person, they prefer iOS TalkBack. Anyway, based on the interface of TalkBack, we provide instruction, turn-by-turn instruction based on indoor map. So this is the last application that I show for this talk.
15:44
Okay, so Inos project is a project for five years, ending at 2021. So we aim to set up an ecosystem around the indoor JML. And yep, so if you are interested
16:00
in working together with me, please contact with me. So we have a lot of flexibility of our project. So okay, and you may find a lot of, so these solution are mostly open source software. So you can find the software on the GitHub of our lab, which is here, right. Okay, that's all.
16:25
Thank you. I kept the time limit. All right, okay. Any questions?
16:41
Indoor mobile navigation for the blind people? Yes. All right. All right. Okay. Yeah, this is very important question.
17:02
So. Sorry? So his question is how we can get the position in indoor space. So for outdoor space, we have a GPS. This is a dominant technology for almost everywhere. But indoor space, it is not the case.
17:21
As you know, GPS is now working in indoor space. So there are many methods. The first one is geomagnetic sensor. Geomagnetic sensor is a very nice tool because we only rely on the geomagnetic signals. So we don't need any infrastructure.
17:40
And the second one is Wi-Fi signal. We have a lot of APs. So this is the second option. Third one is a UWB ultra wide band. Ultra wide band gives a very precise location, but it is quite expensive so far. And the next one is the light-based sensor.
18:01
There are numerous ways. And the final one is called the PDR. Pedestrian dead reckoning. That means if we know the location of, suppose that I turn just from left, then suppose that I know that location. That by counting the number of step, I can estimate my location.
18:22
So this is the, but I'd like to say that it is, first, it is hybrid. No one single method work for every case. And the second, it should be case by case. We suppose that, for example, this place, I believe that geomagnetic sensor work very well.
18:41
But if you go underground, like the subway, there are a power line of high voltage. It is sort of the geomagnetic field. So it doesn't work. So it depends. So if you have the area that you want to apply the indoor positioning, you have to go there
19:00
and you have to test every method to see what is the best solution. And yeah, I'd like to mention one another, which is a beacon. So for example, iBeacon from Apple or Eddystone, these are two types of a beacon which are working very well. Any other question?
19:21
Thank you.
Recommendations
Series of 38 media