Graduation Year

2013

Document Type

Thesis

Degree

M.S.C.S.

Degree Granting Department

Engineering Computer Science

Major Professor

Miguel A. Labrador

Co-Major Professor

Sean J. Barbeau

Keywords

Clustering, GPS, Mobile, POI, Tracking

Abstract

In addition to the emergence of smartphones and tablets in recent years, the rise of Global Navigation Satellite Systems (GNSS) has allowed mobile tracking applications to become popular and be put into many uses. Analyzing tracking records to identify points of interest (POIs) is useful for both prediction applications and research such as human behavior analysis, transportation planning, and especially travel surveys. Past research in travel surveys has shown that a GPS mobile phone-based survey is a useful tool for collecting information about individuals. Moreover, a passive travel survey collection is preferred to an active travel survey method by the respondents and the analysts because it is proven to be less error prone. However, passive collection remains a challenge due to a lack of high accuracy algorithms to automatically identify trip starts and trip ends. While travel surveys need a POI identification algorithm to carry out passive information collection, mobile tracking applications must be careful not to affect the user's battery life, which limits the number of GPS coordinates that can be recorded and therefore affects the accuracies of existing POI identification algorithms. This thesis presents Automatic Spatial Temporal Identification of Points of Interest (ASTIPI), an unsupervised spatial temporal algorithm to identify POIs. ASTIPI utilizes the temporal and spatial properties of the dataset to obtain a high accuracy of POI identification, even on a reduced GPS dataset that uses techniques to conserve battery life on mobile devices. While reducing outliers within POIs, ASTIPI also has a linear running time and maintains the temporal orders of the location data so that arrival and departure information can be easily extracted and thus, users' trips can be quickly identified. Using data from real mobile devices, evaluations of ASTIPI and other existing algorithms are performed, showing that ASTIPI obtains the highest accuracy of POI identification with an average accuracy of 88% when performing on full datasets generated using the GPS Auto-Sleep module and an average accuracy of 59% when performing on a reduced dataset generated using both the GPS Auto-Sleep module and the Critical Points algorithm.

Share

COinS