Using reverse viewshed analysis to assess the location correctness of visually generated vgi
0 likes724 views
This document discusses using reverse viewshed analysis to assess the location correctness of geotagged photos from Flickr. The researchers analyzed photos of two landmarks in Berlin - the Brandenburg Gate and Reichstag. They categorized each photo based on whether the geotag and label were correct based on the viewshed analysis and manual inspection. Their results showed geotags and labels were less accurate for photos further from the landmarks and users with fewer total photos. The researchers propose using additional data like observer height and a weighted credibility score to improve location correctness assessment of visually generated VGI.
Using reverse viewshed analysis to assess the location correctness of visually generated vgi
- 1. Using Reverse Viewshed Analysis to Assess the Location Correctness of Visually Generated VGI Hansi Senaratne | Arne Broering | Tobias Schreck 2013 ESRI International User Conference - GIScience Session 10.07.2013
- 2. Volunteered Geographic Information (VGI) What is VGI? • A special case of UGC • For creating geographic information • Contributors are quite often untrained • May or may not be accurate • Various dedicated Web platforms • flickr- 6.7 billion images • OpenStreetMap - 2.75 billion track points ESRI UC 2013. 2 Use of VGI for disaster relief!
- 3. Quality of VGI Hurricane Sandy: a still from the movie “The Day after Tomorrow” (Twitter) Hurricane Sandy: photo-shopped statue of Liberty with a dramatic storm hovering over it (Twitter) Reichstag: geotagged 6km East of the actual location (flickr) Birkenkopf hill: Overlay of several GPS tracks (Wikipedia) ESRI UC 2013. 3
- 4. E.g. Positional accuracy in Source: Goodchild & Li. (2012) ESRI UC 2013. 4
- 5. E.g. Positional accuracy in Using metadata to assess the location correctness and thereby the credibility of flickr contributors ESRI UC 2013. 5
- 6. What is Credibility? The believability of a source or message, which comprises primarily two dimensions, the trustworthiness and expertise Subjective Objective (Hovland et al. 1953) (Flanagin & Metzger 2008) + Source: http://applemintsoda.wordpress.com/2012/03/19/trustworthiness/ http://www.ianbrodie.com/selling/expertise-driven-selling/ ESRI UC 2013. 6
- 7. Our Approach E.g.: “Brandenburg Gate” , “Berlin” ESRI UC 2013. 7
- 8. Approach (I). POIs in Berlin Brandenburg Gate Reichstag Sample of 100 photos for each POI ESRI UC 2013. 8
- 9. DSM dataset from EuroMaps • IRS-P5 Cartosat-1 in-flight stereo data (2012) • Buildings, vegetation • 5 m post spacing • relative vertical accuracy of 2.5 m with a linear error of 90% (LE90). ESRI UC 2013. Source: German Aerospace Center (DLR) 9
- 10. Approach (II). Reverse Viewshed Calculation Source: http://resources.arcgis.com/en/home/ Parameters Default values OFFSETA 1 OFFSETB 0 AZIMUTH1 0 AZIMUTH2 360 VERT1 90 VERT2 -90 RADIUS1 0 RADIUS2 Infinity A reverse viewhsed determines the visibility of a given target point from many observer points (Fisher 1996)ESRI UC 2013. 10
- 11. Approach (III). Categorising Photos Photo Category Correct Geotag Correct Label a No No b No Yes c Yes No d Yes Yes Reverse viewshed Location correctness Manual inspection Label correctness ESRI UC 2013. 11
- 12. Approach (IV). Photo Categories for Brandenburg Gate ESRI UC 2013. 12
- 13. Approach (IV). Photo Categories for Reichstag ESRI UC 2013. 13
- 14. What do you think? • Contributor # 1 – Average tag count per photo: 15 – Contacts count: 1000 – Total photo count: 68,882 • Contributor # 2 – Average tag count per photo: 3 – Contacts count: 13 – Total photo count: 239 ESRI UC 2013. 14
- 15. … Location correctness of #1 & #2 • Contributor # 1 viewshed • Contributor # 2 viewshed Legend Not Visible Visible Legend Not Visible Visible AvgTagCount/photo: 15 ContactsCount: 1000 TotalPhotoCount: 68,882 AvgTagCount/photo: 3 ContactsCount: 13 TotalPhotoCount: 239 ESRI UC 2013. 15
- 16. Some results we found I Flickr metadata Brandenburg Gate Reichstag a(30%) b(19%) c(11%) d(40%) a(27%) b(11%) c(25%) d(37%) Avg. photo tag count 18 8 13 11 35 12 22 10 Avg. user photo count 19,087 3,852 18,354 5,422 8,136 7,928 9,555 2,618 Avg. user contact count 338 111 134 132 108 141 153 110 Avg. distance to the target 626.5 402.9 299.1 161.6 1,321 735.9 510.5 436.6 The further away users are from the POI, the less accurate they get in geotagging and labeling their photos ESRI UC 2013. 16
- 17. Some results we found II Distance to the target User photo count Photo ta1g7 count Correct geotag/label Incorrect geotag/label Correct geotag/label Incorrect geotag/label Correct geotag/label Incorrect geotag/label
- 18. Issues to think about! • Erroneous geotagging using the map interface • Additional data can improve the viewshed – E.g. height of the observer from surface point – Photos taken from higher levels on buildings ESRI UC 2013. 18
- 19. Future Work • Weighted score scheme for Flickr metadata • User interface for quality aware users • Credibility of text based VGI – Twitter credibility assessment • Based on the information spread • Based on credibility indicators i.e., re-tweets, no. of followers, • …. ESRI UC 2013. 19
- 20. Thank You. Contact: [email protected] http://infovis.uni-konstanz.de/~senaratne/ ESRI UC 2013. 20
- 21. BACKUP SLIDES ESRI UC 2013. 21
- 22. Quality Assessment of VGI – Related Work • Distance based TRUST model – Bishr et al. (2008) • User verification – Goodchild (2009); Coleman (2009) • Image recognition (flickr) - Friedland (2011) • Rating systems (GeoLabel) - Lush et al. (2012) • Proprietary data comparison (OSM) – Haklay (2010); Zielstra et al.(2010) ESRI UC 2013. 22