Computerized trip classification of GPS data Extended Abstract

INTRODUCTION Departments of Transportation (DoT) at the state and federal level constantly log the travel behavior of individuals to help with urban planning and to develop transport models. This is often accomplished by placing a Global Positioning System (GPS) in a user's vehicle and augmenting the electronic data with a travel diary to classify the purpose of each trip. This information is combined with Geographical Information System (GIS) data in the form of geocoded maps that identify locations at given longitudes and latitudes. However, the fact that only a small percentage of the U.S. has been encoded in detail and the coding of locations from this data is generally performed in a manual fashion makes the current strategies for classification a labor intensive process that is also incomplete. Another difficulty stems from the required household travel surveys and travel diaries that are commonly used to collect traffic data for deriving and validating the travel models. Due to the tenacity required by the participant to complete the diary and the high probability or errors and omissions, written logs threaten the validity of the data and any subsequent model. As the use portable Global Positioning Systems (GPS) grows, so does the opportunity for the collection of passive GPS data. Although Wolf, et al. showed that it is possible to classify some trips automatically [13], currently automated processes are not well developed. The primary goal of the work presented in this paper is to automate the collection and classification process and remove the majority of the human element in the trip derivation process by creating an automatic classification framework. This innovative work crosses disciplinary borders by utilizing modern computer technologies to assist urban planners and traffic engineers accomplish their immediate goals. This paper proposes a classification framework to convert passively recorded streams of GPS data from a volunteer driver to a geocoded travel diary from which origin-destination matrices (OD) and travel-time matrices [9], the nuts-and-bolts of traffic and urban planners, can be generated. Typical GPS logs contain information such as location (latitude, longitude, altitude), movement (velocity, heading) and time information. The proposed framework utilizes the latitude, longitude and time stamp components to exploit an adaptive spatial (related to location) and temporal (related to time of day) clustering algorithm to identify a point of interest (POI) which represents a set of trips having the same purpose. The study also investigates perceived regularities ascertained from …