Processing low-frequency floating cardata for transportation applications

Abstract

During the previous decade and onwards, utilizing GPS probe data coming from fleet management systems, as floating car data (FCD), has attracted a lot of attention. However, the particularity of processing this kind of data is that they are relatively sparse in time and space. As a consequence, identifying the followed path between consecutive GPS probes involves high uncertainty. Map-matching process facilitates the transformation of these GPS probe data, which are in raw format, into useful information for transportation applications. Map-matching process is divided into two distinct problems; the map-matching and the path inference. The former identifies possible matches of vehicles’ reported position on the network links (projection points), while the latter one identifies the followed path between each pair of consecutive probes (path identification) and the most likely projections (projection matching). The objective of the present Thesis is to investigate how to improve the outcome of processing low frequency FCD by making use of information that already exists in the available GPS probes database but it is not utilized so far. This is accomplished by developing a dynamic Logit model, learned from data, which captures the behavioral aspects describing drivers’ local route choice procedure. Subsequently, the developed model is incorporated into an existing map-matching algorithm, replacing the currently deterministic path inference approach by a probabilistic one, with an objective to improve its performance, and eventually lead to improved travel time estimations. In addition to the above, an iterative process of building up comprehensive historical travel times’ database for the links of the entire network is designed to expedite the application of a dynamic model. The novelty of the applied approach is that it proposes an overall probabilistic path-inference method that utilizes the dynamic route choice behavior concept in both steps of path-inference process. Finally, the designed overall probabilistic path-inference method is incorporated into an existing map-matching algorithm in order to facilitate both a path-based and a travel time-based assessment of the impacts of this incorporation. The results exhibit that the employment of the route choice behavior concept in the map-matching process has, indeed, an impact on the results.