MediaQ¹ - POI

David Steiner
Ramona Zeller
Johannes Ziegltrum

Introduction

People like watching videos, especially if they show interesting things or locations - or cats.

With MediaQ¹ we will soon have the opportunity to watch videos from all over the world. The problem is that we don’t know what a video shows before we have watched it. It would be much easier for us if we knew beforehand whether the video is showing the place we're looking for or not. Additionally it would be handy if we could just search for a place we're interested in and get a list of videos that have captured it.

Introducing Trajectory-To-POI-Mapping - connecting videos to their underlying geographical information.

This way we have the opportunity to search for certain points of interest (POI) like sights, parks or restaurants and are able to tell exactly when an interesting place is visible in a video.

Example use-cases could be:

Someone is telling you about an amazing place you have never been before. No Problem. Just search for all videos that show exactly this place.
People who travel the world can get an impression of what to expect at certain destinations and check out interesting locations before they get there.
People who can't travel the world can see the greatest places on earth through the eyes of others.

On the one hand we have the video database of MediaQ¹, containing videos and their meta-information like trajectories, trajectory points or the perspective. On the other hand we have services like the Google Places API² that returns information about places (establishments, geographic locations, or prominent points of interest). The idea is to connect the information of both systems to a single, even more powerful, application.

Queries and algorithms

1) Find all videos that show a given POI

Naive approach

Foreach video in ALL videos
  trajectory = video.getTrajectory();
  Foreach trajectory_point in trajectory
    if POI is visible at the trajectory_point
      -> add video to result set

"Our" approach using the R-tree index structure

Basis of this query is a R-tree⁷ filled with all videos of the MediaQ database. In order to do this we calculate for each video the minimum bounded rectangel (MBR) of the video trajectory and insert it into the R-tree⁷.

This way we can search now for videos in a certain range on the map.

Filtering with R-tree⁷

First of all we determine the coordinates (latitude, longitude) of the POI via Google Places API². By means of this coordinates and the visibility range we define our search area.

With the MBR of this search area we can run a range query on the R-tree⁷. The result set of this query contains all videos which might be showing the given POI.

Refinement

In the filtering phase we determined the video candidates. Now we have to check for each candidate if the POI is visible in any trajectory point of this video. Therefore we calculate the distance and the angle between each trajectory point of a video and the given POI.

Each trajectory point contains the information about its perspective.

After comparing this values we know exactly if a video is showing the POI we are searching for.

Foreach video in candidates
  trajectory = video.getTrajectory();
  Foreach trajectory_point in trajectory
    calculate the distance and the angle between the trajectory_point and the given POI
    if (distance <= radius) AND (thetax - alpha/2 <= angle <= thetax + alpha/2)
      -> add video to result set

"Our" approach using a Google Document Index

In this approach we use the Google Search API³ which already provides geospatial index structures. However, we have to store the videos as documents in order to use this features. We store the circular approximation of each video trajectory as document and trust Google enough to index them efficiently ;-) The middle point of the circle is saved as geopoint field ("centerPoint") and the search radius ("searchRange") of the trajectory is stored as number field.

The query is solved using a filter step. The maximum search radius of the indexed trajectories is used as upper bound. All videos that have a less or equal distance to the poi are handled as candidates:

String queryString = "distance(centerPoint, " + geoPoint(poiLocation) + ") <= " + maxSearchRange;
Build query with the queryString
Execute the query and add results to candidates list

The refinement step is analogous to the R-tree refinement.

While realizing this query with the Google Search API³ we faced some issues with it:

Only very basic queries are supported by the API (see also http://stackoverflow.com/questions/28213499/is-it-possible-to-use-a-document-field-as-radius-with-distance-function-in-googl)
Data is not stored in the data store but as document
Some functions cannot be used in the local development server (see also https://cloud.google.com/appengine/docs/java/search/devserver)
You have no insight into the index structure

2) Find all POIs that are shown in a given video

The naive approach here would be to iterate through all trajectory points of the video and check ALL POIs if they are visible at each point. Of course, this is not very practical and hard to realize. Maybe Chuck Norris is the only one to get all POIs of the whole world in one single request from the Google Search API³ ;-)

As Chuck Norris was unfortunately not willing to participate in our project, we had to find another way to solve this query.

Filtering with Google Places API²

First of all we calculate the circle that surrounds the trajectory of the given video and add the visibility range. Google Places API² now provides us with the POIs in this search area.

-> Results in POI candidates which might be shown in the video

Refinement

For each trajectory point of the video determine which of the POI candidates is visible. Again we know the perspective of each trajectory point and calculate the distance and the angle to each POI candidate. After comparing this values we get the result set with all visible POIs in this video.

trajectory = video.getTrajectory();
Foreach trajectory_point in trajectory
  Foreach POI in candidates
    calculate the distance and the angle between the trajectory_point and the POI
    if (distance <= radius) AND (thetax - alpha/2 <= angle <= thetax + alpha/2)
      -> add POI to result set

3) Find all videos that are recorded in a certain map section

To reduce the data load and achieve a good performance we decided to load only those videos that are located in the current map section that the user has navigated to.

Naive approach

foreach video in ALL videos
  trajectory = video.getTrajectory()
  get minimal bounding rectangle (MBR) of trajectory
  if MBR intersects with query window (range)
    foreach trajectory_point in trajectory
       if range contains trajectory_point
         -> add video to result set
         break;

"Our" approach using the R-tree index structure

The R-tree can be used directly to find videos in a certain area.

"Our" approach using a Google Document Index

We additionally store the minimal bounding rectangle (MBR) of the video trajectory in the document. The MBR is represented by two opposite corner points. The location with the minimum longitude and minimum latitude is stored as geopoint field ("minPoint") and the location with the maximum coordinates finds oneself also as geopoint field ("maxPoint") in the document.

In the refinement step, all videos that have either their minimum point or their maximum point lying in the circumcircle of the range are loaded as candidates. The circumcircle of the range has to be used because the Google Search API³ does not support queries for intersections with geometrical figures. It only lets you query for distances between geopoints.

middle = center of circumcircle around rectangular range
radius = distance between middle and one corner point
String queryString = "distance(minPoint, " + geoPoint(middle) + ") <= " + radius;
     queryString += " OR distance(maxPoint, " + geoPoint(middle) + ") <= " + radius;
Build query with the queryString
Execute the query and add results to candidates list

In the refinement step the candidates are checked if they have at least one trajectory point that is located in the range area:

foreach video in candidates
  trajectory = video.getTrajectory()
  get minimal bounding rectangle (MBR) of trajectory (minPoint, maxPoint)
  if MBR intersects with query window (range)
    foreach trajectory_point in trajectory
       if range contains trajectory_point
         -> add video to result set
         break;

For the intersection test of the range with the MBR of the trajectory the LatLonRect⁸ class of the unidataCommon library is used.

Backend

####Get all videos within certain bounds:

Request:

http://mediaq-poi.appspot.com/videos?
action=range_query&
bound1_lat=48.158445160282696&bound1_lng=11.624752960540718&
bound2_lat=48.142611618392216&bound2_lng=11.565401039459175

Parameters: North-East and South-West Latitude and Longitude.

Returns: A list of videos.

Response:

{
  status: "ok",
  videos: [
    {
      fileName: "3h5yxvywo2kg_2014_12_8_Videotake_1418038378126.mp4",
      filePath: "http://mediaq.dbs.ifi.lmu.de/MediaQ_MVC_V2/video_content/3h5yxvywo2kg_2014_12_8_Videotake_1418038378126.mp4",
      id: "3h5yxvywo2kg_2014_12_8_Videotake_1418038378126.mp4",
      latitude: 48.152392,
      longitude: 11.591949
    },
    {
      fileName: "-z0waj48xtiyf_2014_12_8_Videotake_1418038301552.mp4",
      filePath: "http://mediaq.dbs.ifi.lmu.de/MediaQ_MVC_V2/video_content/-z0waj48xtiyf_2014_12_8_Videotake_1418038301552.mp4",
      id: "-z0waj48xtiyf_2014_12_8_Videotake_1418038301552.mp4",
      latitude: 48.152059,
      longitude: 11.592511
    },
    ...
  ]
}

####Get all POIs for a video

Request:

http://mediaq-poi.appspot.com/video/3h5yxvywo2kg_2014_12_8_Videotake_1418038378126.mp4

Parameters: The video id.

Returns:

The video details
The calculated center of the videos trajectory
The radius enclosing the trajectory that has been used for the range query on Google Places API²
A list of nearby POIs - the candidates received from Google Places API²
A list of visible POIs - calculated by our algorithm
A list of user positions with latitude, longitude and viewing direction sorted by position in the video (in seconds)
The timeline sorted by position in the video (in seconds) describing what POI is visible at this time

Response:

{
  center: {
    latitude: 48.1523895,
    longitude: 11.592268
  },
  searchRange: 150,
  visiblePois: [
    {
      reference: "CpQBgwAAAOYnaYrxXr71LZvu-LQXNkn9C2y7FJ3PzsubGEYLZyMziA5xhofRcD3qo60c-5uHEyVySlGsfLu8BlMqJ3SJYnIEv5Q9ft-73vemoVO_KQGltSTQ3J87vqRL_UlcmhGPLI7hGoa-2LiS6MLBaQGABWY577Gls81p8RHY91Eti8hYjbjc5sHkW7IapKn6ndpriBIQWVZZFJr0vDokKevMuBxZcxoUcWLpwYNfNSH_O9puzuwRWJZujSk",
      id: "ChIJSynAOJZ1nkcRLbbGZPaCR1o",
      latitude: 48.152005,
      longitude: 11.592346,
      name: "Christmas Market at Chinese Tower"
    },
    ...
  ],
  nearbyPois: [
    {
      reference: "CoQBegAAAPnqBhtWfnaD_ib0RkV-znyJyWpQiN9rGonzjQgKLoSplnH-q7WMSLWIvvAHihFTIGSH-4bMpQqraGZQVu2MXRZqhdwJAzUXPUIxNMm-TgaQupimLuyCZ9HoN_P2u7P0eFm7i8II4ZXY_iLSVjsgXA8JuK9UISOEWeZKDmVnpC4cEhDtevrh-YUOq-Vg4SXKA8RMGhT2Rd8n2o-U86lQNdZqacDVCICbRg",
      id: "ChIJayv4lZd1nkcR0e_vfGLfm8k",
      latitude: 48.16423229999999,
      longitude: 11.6055522,
      name: "English Garden"
    },
    ...
  ],
  posTimeline: {
    0: {
      latitude: 48.152392,
      longitude: 11.591949,
      thetaX: 259.97482
    },
    ...
  },
  timeline: {
    0: [ ],
    1: [ ],
    2: [
      {
        reference: "CpQBgwAAAOYnaYrxXr71LZvu-LQXNkn9C2y7FJ3PzsubGEYLZyMziA5xhofRcD3qo60c-5uHEyVySlGsfLu8BlMqJ3SJYnIEv5Q9ft-73vemoVO_KQGltSTQ3J87vqRL_UlcmhGPLI7hGoa-2LiS6MLBaQGABWY577Gls81p8RHY91Eti8hYjbjc5sHkW7IapKn6ndpriBIQWVZZFJr0vDokKevMuBxZcxoUcWLpwYNfNSH_O9puzuwRWJZujSk",
        id: "ChIJSynAOJZ1nkcRLbbGZPaCR1o",
        latitude: 48.152005,
        longitude: 11.592346,
        name: "Christmas Market at Chinese Tower"
      },
      {
        reference: "CoQBewAAAJif1b-ypdFUx1xCd2HMd9yU6Kjih1aJ71j_frWYGXHhCgS17EH2TLxT6cf3ZNnSsvzGGoV6P4XKPed9rTXs4TVj6eVR4uzVmjK805ZKwAbVGbXDeAM_JnGX-qePrsKJrD9EOlZ1OtNlQF22P7L48AbOqy5vnDlKaQQ4V5fo4mBkEhAPeH6opDtO2hQqxjCATZ7QGhTB-SMddeyJc47Y4SpkMrZKM9y-mQ",
        id: "ChIJx8O3wZd1nkcRSD11Hq3mOm0",
        latitude: 48.152005,
        longitude: 11.592346,
        name: "Ochsenbraterei Haberl OHG"
      }
    ],
    ...
  },
  video: {
    fileName: "3h5yxvywo2kg_2014_12_8_Videotake_1418038378126.mp4",
    filePath: "http://mediaq.dbs.ifi.lmu.de/MediaQ_MVC_V2/video_content/3h5yxvywo2kg_2014_12_8_Videotake_1418038378126.mp4",
    id: "3h5yxvywo2kg_2014_12_8_Videotake_1418038378126.mp4",
    latitude: 48.152392,
    longitude: 11.591949
  }
}

####Get all videos for a POI

Request:

http://mediaq-poi.appspot.com/poi/ChIJ6Q6XOph1nkcRQWtXFc8qRRg

Parameters: The POI id.

Returns: The POI details and a list of videos that show the POI.

Response:

{
  poi: {
    reference: "CnRqAAAAd46UWaGluNFcTOmF4ex9QbKO6xE52f6fI7RU3jD69KckanXnp-CtKxi7R0YagbEs-GC99Xbc-Xpe4wOgGY1P2Rq_2fWZkTA76o8auxRHF7XjiJuDnnqd9FdUi5fgZmSgsbJ4t8BOFfywXmwaw8IozBIQNWIvBRpugSfnEJ2g7IR_sBoUkt2fWdwj_XP3P7TiQpTkhondwP0",
    id: "ChIJ6Q6XOph1nkcRQWtXFc8qRRg",
    latitude: 48.152556,
    longitude: 11.592084,
    name: "Chinesischer Turm"
  },
  videos: [
    {
      fileName: "3h5yxvywo2kg_2014_12_8_Videotake_1418038378126.mp4",
      filePath: "http://mediaq.dbs.ifi.lmu.de/MediaQ_MVC_V2/video_content/3h5yxvywo2kg_2014_12_8_Videotake_1418038378126.mp4",
      id: "3h5yxvywo2kg_2014_12_8_Videotake_1418038378126.mp4",
      latitude: 48.152392,
      longitude: 11.591949
    },
    {
      fileName: "3h5yxvywo2kg_2014_12_8_Videotake_1418037868170.mp4",
      filePath: "http://mediaq.dbs.ifi.lmu.de/MediaQ_MVC_V2/video_content/3h5yxvywo2kg_2014_12_8_Videotake_1418037868170.mp4",
      id: "3h5yxvywo2kg_2014_12_8_Videotake_1418037868170.mp4",
      latitude: 48.152499,
      longitude: 11.592498
    },
    ..
  ]
}

Frontend

To demonstrate the abilities of our implementation we developed a sample application that queries our interface.

You can check it out here.

The application features a map that displays all videos in the visible bounds. If you click on a video-marker the video details are loaded. Once this is done the video is played and the trajectory shown. While the video is playing the current location and the viewing direction as well as the visible POIs at the current time keep being updated and displayed on the map.

It is also possible to search for a POI using the search field in the header. The search is using the autocomplete function provided by the Google Places API². When a POI is selected it's details are loaded and the POI and the videos that show it are displayed on the map.

Technical documentation

Google Places API²

Used for getting a list of POIs in a given range, the candidates for our algorithm, and searching for POIs.
Google App Engine³

The frontend and backend are hosted on Google App Engine³. App Engine is a cloud service for hosting Java, Python, PHP or Go applications. It also features "a schemaless NoSQL datastore providing robust, scalable storage" which we use for persisting the videos, trajectories and trajectory points.
Google Search API⁴

Used for indexing the videos by their circular approximation and for getting the candidate videos for a given POI.
AngularJS⁵

Angular is a framework for JavaScript with features like two-way data-binding and dependency injection that allowed us to write a well structured frontend application.
GruntJS⁶

Grunt⁶, The JavaScript Task Runner is automatically building our frontend distribution - including JavaScript and image minification and much more.
R-tree⁷

A high performance Java version of the R-tree⁷ spatial indexing algorithm.

UML

Use cases of the mediaq-poi application

Sequence diagram to show how the query "get videos for POI" is solved. The servlet receives the request with the POI's place id as parameter. It hands over the place Id to the PoiService which retrieves the place details via the Google Places Api and returns the according Poi object. The servlet then can call the PoiService again to retrieve the videos that record the geo position of the POI. Depending on the set algorithm approach type (RTREE, GOOGLE_DOCUMENT_INDEX or NAIVE) the PoiService instance gets the video results and returns them to the servlet. Finally, the servlet puts the video data as JSON into the http response and returns it to the frontend where the data will be further handled and shown to the user.

Sequence diagram to show how the query "get POIs for video" is solved. The servlet receives the request with the video's id as parameter. It loads the Video object from the datastore (not shown in sequence diagram) and calls the getVisiblePois()-method of the PoiService. The service now gets the circular approximation of the video's trajectory and requests the Google Places API² to return all places within this circular range. All this places are handled as candidates. In the refinement step all trajectory points are iterated and for every single point all candidate POIs are checked if they are visible at the point, If yes, they are added to the result lists. In the end, the results are returned to the servlet which puts them together with other information about the video to the http response (as JSON). The frontend can then display the information.

Simplified class diagram of mediaq-poi application. Only the main relations are shown and the entity classes are not specified in detail. The servlets mainly communicate with the IPoiService interface so that the business logic is capsulized. The entity classes are very simple data containers that are used for communication between the different components.

Performance Evaluation

In order to measure the time that our different algorithms are consuming we integrated the StopWatch⁹ provided by the Apache commons-lang3 library.

 StopWatch stopWatch = new StopWatch();
 stopWatch.start();
 // query with {NAIVE, RTREE, GOOGLE_DOCUMENT_INDEX}
 stopWatch.stop();
 long elapsedTimeInMilliseconds = stopWatch.getTime()

Below the performance evaluation for the three different approaches NAIVE, RTREE and DOCUMENT_GOOGLE_INDEX.

1) POI query for 'Chinesicher Turm'

NAIVE:

/poi/ChIJ6Q6XOph1nkcRQWtXFc8qRRg
	PoiServlet getPoiDetails: Getting details for poi id ChIJ6Q6XOph1nkcRQWtXFc8qRRg
	PoiService getVideosNaive: Performing video query for location with naive approach
	PoiService getVideosNaive: Found 46 videos for the given location (naive approach)
	PoiService getVideos: Got videos for location in 1574 milliseconds (using NAIVE)

RTREE:

/poi/ChIJ6Q6XOph1nkcRQWtXFc8qRRg
	PoiServlet getPoiDetails: Getting details for poi id ChIJ6Q6XOph1nkcRQWtXFc8qRRg
  	VideoRTree getInstance: Getting VideoRTree instance
  	VideoRTree getCandidates: Found 58 candidates for geo location in the r tree
   	VideoRTree getVideos: Found 46 videos for geo location in the r tree
   	PoiService getVideos: Got videos for location in 47 milliseconds (using RTREE)

GOOGLE_DOCUMENT_INDEX:

/poi/ChIJ6Q6XOph1nkcRQWtXFc8qRRg
	PoiServlet getPoiDetails: Getting details for poi id ChIJsT-BEJd1nkcRWL5StFHCk5k
	PersistenceFacade getVideos: Found 46 video for the given location (using Google document index)
	PoiService getVideos: Got videos for location in 640 milliseconds (using GOOGLE_DOCUMENT_INDEX)

Query was executed 5 times and mean was calculated:

RTREE (47 milliseconds) < GOOGLE_DOCUMENT_INDEX (640 milliseconds) < NAIVE (1574 milliseconds)

2) POI query for 'Monopteros'

NAIVE:

/poi/ChIJsT-BEJd1nkcRWL5StFHCk5k
	PoiServlet getPoiDetails: Getting details for poi id ChIJ6Q6XOph1nkcRQWtXFc8qRRg
	PoiService getVideosNaive: Performing video query for location with naive approach
	PoiService getVideosNaive: Found 1 video for the given location (naive approach)
	PoiService getVideos: Got videos for location in 1700 milliseconds (using NAIVE)

RTREE:

/poi/ChIJsT-BEJd1nkcRWL5StFHCk5k
	PoiServlet getPoiDetails: Getting details for poi id ChIJ6Q6XOph1nkcRQWtXFc8qRRg
  	VideoRTree getInstance: Getting VideoRTree instance
   	VideoRTree getCandidates: Found 5 candidates for geo location in the r tree
   	VideoRTree getVideos: Found 1 video for geo location in the r tree
   	PoiService getVideos: Got videos for location in 6 milliseconds (using RTREE)

GOOGLE_DOCUMENT_INDEX:

/poi/ChIJsT-BEJd1nkcRWL5StFHCk5k
	PoiServlet getPoiDetails: Getting details for poi id ChIJsT-BEJd1nkcRWL5StFHCk5k
	PersistenceFacade getVideos: Found 1 video for the given location (using Google document index)
	PoiService getVideos: Got videos for location in 49 milliseconds (using GOOGLE_DOCUMENT_INDEX)

Query was executed 5 times and mean was calculated:

RTREE (6 milliseconds) < GOOGLE_DOCUMENT_INDEX (49 milliseconds) << NAIVE (1700 milliseconds)

3) Initial range query at 'Chinesischer Turm'

NAIVE:

/videos?action=range_query&bound1_lat=48.16840282484953&bound1_lng=11.61478721537776&bound2_lat=48.14598681662252&bound2_lng=11.555435294296217
	PoiService getVideosInRangeNaive: Performing range query with naive approach
	PoiService getVideosInRangeNaive: Found 140 videos in range (naive approach)
	PoiService getVideosInRange: Got videos in range in 2102 milliseconds (using NAIVE)

RTREE:

/videos?action=range_query&bound1_lat=48.16376962982234&bound1_lng=11.631125889160103&bound2_lat=48.137284953024235&bound2_lng=11.55902811083979
	VideoRTree getInstance: Getting VideoRTree instance
	VideoRTree getVideosForArea: Found 140 videos for range in the r tree
	PoiService getVideosInRange: Got videos in range in 0 milliseconds (using RTREE)

Query was executed 5 times and mean was calculated:

RTREE (0 milliseconds) << NAIVE (2102 milliseconds) < GOOGLE_DOCUMENT_INDEX (2180 milliseconds)

4) Range query at 'Freising'

NAIVE:

/videos?action=range_query&bound1_lat=48.41605498486009&bound1_lng=11.777233489160153&bound2_lat=48.38970080144271&bound2_lng=11.70513571083984
	PoiService getVideosInRangeNaive: Performing range query with naive approach
	PoiService getVideosInRangeNaive: Found 5 videos in range (naive approach)
	PoiService getVideosInRange: Got videos in range in 1649 milliseconds (using NAIVE)

RTREE:

/videos?action=range_query&bound1_lat=48.41605498486009&bound1_lng=11.777233489160153&bound2_lat=48.38970080144271&bound2_lng=11.70513571083984
	VideoRTree getInstance: Getting VideoRTree instance
	VideoRTree getVideosForArea: Found 5 videos for range in the r tree
	PoiService getVideosInRange: Got videos in range in 0 milliseconds (using RTREE)

Query was executed 5 times and mean was calculated:

RTREE (0 milliseconds) << NAIVE (1649 milliseconds) < GOOGLE_DOCUMENT_INDEX (1833 milliseconds)

Downloads

Presentation (in German on Google Drive)
Sourcecode (on Github)

References

1: MediaQ http://mediaq.usc.edu/
2: Google Places API https://developers.google.com/places/documentation/
3: Google App Engine https://cloud.google.com/appengine/docs
4: Google Search Api https://cloud.google.com/appengine/docs/java/search/
5: AngularJS https://angularjs.org/
6: GruntJS http://gruntjs.com/
7: R-tree https://github.com/aled/jsi/
8: LatLonRect http://www.unidata.ucar.edu/software/thredds/v4.3/netcdf-java/v4.3/javadoc/ucar/unidata/geoloc/LatLonRect.html
9: StopWatch https://commons.apache.org/proper/commons-lang/javadocs/api-3.1/index.html?org/apache/commons/lang3/time/StopWatch.html

Developed as part of MediaQ: Practical Seminar on Big Data for Social Media at Ludwig Maximilians University Munich, Insitute of Computer Science by David Steiner, Ramona Zeller and Johannes Ziegltrum.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

README.md

README.md

MediaQ¹ - POI

Introduction

Queries and algorithms

1) Find all videos that show a given POI

Naive approach

"Our" approach using the R-tree index structure

Filtering with R-tree⁷

Refinement

"Our" approach using a Google Document Index

2) Find all POIs that are shown in a given video

Filtering with Google Places API²

Refinement

3) Find all videos that are recorded in a certain map section

Naive approach

"Our" approach using the R-tree index structure

"Our" approach using a Google Document Index

Backend

Frontend

Technical documentation

Performance Evaluation

1) POI query for 'Chinesicher Turm'

2) POI query for 'Monopteros'

3) Initial range query at 'Chinesischer Turm'

4) Range query at 'Freising'

Downloads

References

Files

README.md

Latest commit

History

README.md

File metadata and controls

MediaQ1 - POI

Introduction

Queries and algorithms

1) Find all videos that show a given POI

Naive approach

"Our" approach using the R-tree index structure

Filtering with R-tree7

Refinement

"Our" approach using a Google Document Index

2) Find all POIs that are shown in a given video

Filtering with Google Places API2

Refinement

3) Find all videos that are recorded in a certain map section

Naive approach

"Our" approach using the R-tree index structure

"Our" approach using a Google Document Index

Backend

Frontend

Technical documentation

Performance Evaluation

1) POI query for 'Chinesicher Turm'

2) POI query for 'Monopteros'

3) Initial range query at 'Chinesischer Turm'

4) Range query at 'Freising'

Downloads

References

MediaQ¹ - POI

Filtering with R-tree⁷

Filtering with Google Places API²