I. Introduction
In recent years, naturalistic driving studies (NDS), including naturalistic field operational tests, have been employed all around the globe, providing an important source of data for analysis and enabling a better understanding of driver behavior and traffic safety, for example 100-car [1], [2] and SHRP2 in the U.S.A. [3], [4], euroFOT [5], PROLOGUE [6], and UDRIVE [7] in Europe, as well as NDS in Australia [8] and in Japan [9]. In NDS, data is collected automatically for all driving sessions in a large fleet of vehicles for several months. These automatic recordings include vehicle data such as speed and direction; environmental conditions, lane position, location and surrounding traffic recorded by radar, video and other external instrumentation; and video recordings of the drivers face, pedal, and eye movements. The data provided by the NDS design thus offer many opportunities for analysis of both normal driving and safety critical events, and is richer than more traditional data sources such as crash databases [10], [11].