Longitudinal driving behavior: Theory and empirics

Congestion is a serious problem in many countries around the world. Consequently a lot of effort is put in inventing smart methods for reducing congestion. Whether these measures lead to the desired effect appears to be largely dependent on the driving behavior of individual road users. It is therefore not surprising that the effects of measures are often predicted in advance using simulation tools in which the movements of all vehicles on given roadway stretches are simulated separately using dedicated mathematical models. These predictions are of course only reliable when the assumptions about the driving behavior of individual drivers are realistic. To increase the insights into the behavior of drivers and to improve mathematical models describing this behavior we analyze in this thesis how drivers react to vehicles in front, i.e. we analyze how drivers react to changes in the dynamics of vehicles in front, which distance they want to keep and so on. To get a complete view of this so-called longitudinal driving behavior we use image sequences collected using a digital camera attached to a helicopter. We use these images to derive the dynamics (like positions and speeds) of all drivers driving at the observed roadway stretch. The resulting trajectory data are input to an automated calibration procedure with the aim to calibrate a broad range of mathematical models making different assumptions about human driving behavior. These calibrations are performed for all observed drivers separately. The properties of the applied calibration procedure are studied in this thesis. We, firstly, use the calibration results to analyze the degree of heterogeneity present in longitudinal driving behavior. More specific, by comparing model performances and estimated parameter values between drivers we quantify differences in the driving behavior of individual drivers. In our empirical analyses on heterogeneity within the group of drivers of person cars, we show, for example, that the driving styles of drivers differ considerably. Clear differences are identified between the speed-dependent distances drivers want to keep to the driver in front of them. Also the importance person car drivers attach to actually reaching this distance appears to be driver dependent. In our analyses on differences between the behaviors of truck drivers and person car drivers, we show that truck drivers in general appear to drive with a more constant speed than person car drivers. Person car drivers turn furthermore out to be more eager in restoring large deviations from their desired distance than truck drivers. Secondly, we use the calibration results to examine to how many vehicles in front a driver reacts. Although it has often been assumed that drivers look further ahead than their first leader, no empirical evidence has been provided so far. In this thesis, we show that more than half of the considered person car drivers looks further ahead than their direct leader. At least 20% appears to consider even more than two direct leaders. Especially the relative speed regarding direct leaders further downstream turns out to be of influence to the dynamics of the following car. We also show that the number of leaders considered differs between drivers. Even when drivers consider the same leaders it appears that the extents to which these leaders actually influence the longitudinal behavior of the following vehicle are strongly driver/vehicle combination dependent. To determine the effects of our empirical findings on traffic flow predictions we implement them in a dedicated simulation tool. The simulation results indicate clearly that model predictions are very sensitive to the behavioral assumptions (like the assumed degree of heterogeneity) made in the simulation tool. This again stresses the importance of our empirical analyses.Trai