Monday, June 04, 2007

GPS data

Over the past two years, Microsoft has been gathering GPS data from volunteer drivers. Each driver has one of our recording GPS receivers on their dashboard for two weeks. So far, we have data from 227 drivers, comprising over 1.7 million time-stamped (latitude, longitude) points. This talk will give a brief overview of several different research projects we have completed based on this data. These projects include:

* Destination Modeling -- We characterize where and when people go and how quickly they fall into a rut of visiting very few new destinations. * Predestination -- We use our destination models to predict where a driver is driving as the trip progresses. * Snap-To-Road -- Inferring which road a driver is on is harder than just snapping to the nearest one. Our snap-to-road algorithm improves by accounting for the time stamps of the GPS data. * Personalized Routes -- We can plan routes that drivers prefer by looking at where they've already driven. * Location Privacy -- Can we protect the privacy of our volunteer subjects by simply anonymizing their data? We show it is possible to infer the identities of some of them from anonymized GPS data, and we demonstrate different strategies for protecting their privacy by corrupting the data.

No comments: