Abstract:
|
There is considerable interest in characterizing similarities and differences in the behavior of drivers with a primary motivation being safety considerations. Although naturalistic driving studies collect continuous and high resolution measurements of driver's speed and range (distance between adjacent vehicles), methods to analyze such data are underdeveloped. This project examines several exploratory tools for characterizing similarities (and differences) among drivers based on data on speed and range to a leading vehicle. We propose a novel method based on the earth mover's distance (EMD), also called Mallow's distance or Wasserstein metric, to characterize the level of similarity between drivers in the study. The performance of EMD is studied and compared to several other metrics. Our findings show that the EMD has a number of good properties. For instance, the distance structure is robust to initial parameter settings and can be better approximated by two dimensional Euclidean distance. A few examples will be used to illustrate its properties and advantages.
|