iDiary: A Searchable Diary of GPS Data

Applications

The Inventors have developed iDiary, a patent-pending system that turns large GPS signals collected from smart-phones into textual descriptions of the trajectories. The system features a user interface similar to Google Search that allows users to type text queries on their activities (e.g. “Where did I buy books?”) and receive textual answers based on GPS signals. This technology can help to track urban movement patterns to better model and forecast global activities in a community—for example, to understand spatial trends (where people are aggregating), temporal trends (when people go where) and other urban characteristics (e.g. speed of traffic flow, travel time, etc.)

Problem Addressed

Trends in population movement/mobility are important aspects of urban life that can motivate new predictive models of population activity at the individual or community level. However, mining large amounts of GPS data into useful, human-readable information is difficult due to the relative lack of prior work and the UI of existing GPS applications that target map-navigation over text search. The Inventors take on this challenge with “iDiary,” a novel patent-pending system that processes data collected from a sensor network of mobile phones, extracts the semantic meaning of a user/community’s main activities and trends, and provides a text-based UI that allows easy search and management of this information. 

Technology

Once a user installs iDiary on their iPhone, the application collects GPS-points (traces/signal) and transmits them to the Inventors’ lab server. This is done in the background using location services of the iPhone. Since these services consume large amount of power, the GPS such off when the user is stationary for ten minutes. When the server receives the user’s GPS points, it compresses them for efficient storage and to speed up the text-search algorithm. The linear simplification of the signal is used for constructed its semantic compression (coreset) that consists of the k-segment mean with a “smart” non-uniform sample of weighted representative GPS-points. Coresets are calculated in a streaming setting to result in a constant-sized coreset for a user’s data that is continually updated with new data points.

 

When the user runs a text query search on their iPhone, the terms in the query are translated into a (sparse) vector of code words. This vector is projected on the semantic space and replaced by its (dense) semantic j-vector of activities. On the semantic space, the application computes the similar documents (closet vectors) to the query and returns them.

Advantages

  • First system to provide intuitive, text-based search and retrieval of GPS-based information without having to rely on maps or databases.
  • Algorithms for semantic compression and trajectory clustering of massive GPS-signals compute critical locations of a user within provable error bounds.