Exploiting User Generated Content for Mountain Peak Detection

1/05/2013

SoHuman 2013 atACM Web Science 2013 1

Exploiting User Generated Content for Mountain Peak Detection

Roman Fedrov, Davide Martinenghi, Marco Tagliasacchi,

Andrea Castelletti

2

Passive Human Computation

User Effort

SoHuman 2013 atACM Web Science 20131/05/2013

3

Passive Human Computation

Collective Intelligence

SoHuman 2013 atACM Web Science 20131/05/2013

Our GoalRetrieve the collections of the mountain appearances in different time instants and build environmental models.

201320122011201020092008

Key ProblemIdentify the mountains

4

Key Problem: Object Identification

Input

SoHuman 2013 atACM Web Science 20131/05/2013

46° 0’ 48.51” N7° 48’ 6.62” E

http://www.udeuschle.de/Panoramen.html

5

Key Problem: Object Identification

Output

SoHuman 2013 atACM Web Science 20131/05/2013

6

Matching Algorithm

SoHuman 2013 atACM Web Science 20131/05/2013

StepsScaleRenderExtract EdgesFilterMatch

7

Matching Algorithm

SoHuman 2013 atACM Web Science 20131/05/2013

StepsScaleRenderExtract EdgesFilterMatch

8

Matching Algorithm

SoHuman 2013 atACM Web Science 20131/05/2013

StepsScaleRenderExtract EdgesFilterMatch

9

Matching Algorithm

SoHuman 2013 atACM Web Science 20131/05/2013

StepsScaleRenderExtract EdgesFilterMatch

10

Matching Algorithm

SoHuman 2013 atACM Web Science 20131/05/2013

StepsScaleRenderExtract EdgesFilterMatch

Adapted from matching technique by Baboud et al. “Automatic photo-to-terrain alignment for the annotation of mountain pictures”, CVPR 2011

11

Results

SoHuman 2013 atACM Web Science 20131/05/2013

62% of correct match as maximum result

81% of correct match in top-10 positions

12

Potential Applications

SoHuman 2013 atACM Web Science 20131/05/2013

http://www.nohrsc.noaa.gov/

Augmented Reality

Environmental Models

Colle dei BreuilFurgggrat