Personal tools
 

Linear Decompositions of Long Term Time-Lapse Imagery Are Surprisingly Informative

— filed under:

Robert Pless, Washington University in St. Louis

What
  • Computer Science Seminar
When Tue, May 06, 2014
from 11:00 AM to 11:50 AM
Where TBA
Add event to calendar vCal
iCal
Part of the 2014 Case Lecture.

The web has an enormous collection of live cameras view image parks, roads, cities, beaches, mountains, ski-resorts, buildings and more. Over the last 5 years, I have been archiving imagery from most (>25000) publically available outdoor cameras, and working to understand how to effectively use this massively distributed resource as a tool for phenology, environmental and atmospheric measurement.

Because these cameras are fixed in place and watch the same scene over time, this set of images is highly structured.  Because of this structure, simple algorithms based on PCA can be used for diverse purposes such as geo-locating where in the world a camera is, solving for the 3D structure of the scene in view, or segmenting the scene into component parts.  With extra information for each image such as the wind direction or sun direction, then methods based on Canonical Correlation Analysis can solve for the relative orientation of the camera, and the surface normal of parts of the scene.  I will conclude with some of the challenges we continue to face in working with these half-a-billion images.

« September 2017 »
September
SuMoTuWeThFrSa
12
3456789
10111213141516
17181920212223
24252627282930
Upcoming Events
Computer Science Seminar
Tue, Sep 26, 2017
An Introduction to Persistent Homology by David Letscher, SLU
Math/CS Club
Wed, Sep 27, 2017
Chasing the Sun Kevin Scannell, SLU
Algebra Seminar
Thu, Sep 28, 2017
An introduction to cluster superalgebras 4 Jimmy Mixco, Saint Louis University
Previous events…
Upcoming events…