Addressing Some Issues of Map-Matching for Large-Scale, High-Frequency GPS Data Sets

Data from GPS-enabled vehicles has become more and more widely used by travel-behavior researchers and transportation system modelers. However as the GPS data has measurement and sampling errors it becomes a non-trivial task to infer a map feature associated with a sequence of GPS measurements, especially as maps features may also have inaccuracies. The task of assigning a set of GPS points to a set of map features is called the map matching problem, and the requirement for the assigned features to form a consistent travel route adds additional complexity. The majority of the existing algorithms concentrate on scenarios when sampling rate is low and/or measurement error is high. However, as GPS devices become more accurate and sampling rate becomes higher, a new issue arises, the issue of efficiently of analyzing large scale high frequency GPS data sets. In this paper the authors analyze a high-accuracy and high-frequency GPS data set collected from instrumented vehicles that participated in Safety Pilot Model Deployment project using a modified version of the Multiple Hypothesis Technique to match network links, with accuracy and computational efficiency being the focus. The authors build on previous work in several ways: (i) the authors proposed a speed-up step that significantly reduces the number of candidate paths, (ii) the authors improved the way GPS trace segments are matched to road network at turns and intersections and(iii) the authors added new filtering step to identify U-Turn movements. In addition to these improvements, the authors demonstrate the process of manually fitting the parameters of the algorithm and suggest future direction on how the process of parameter training can be automated.