This paper first reports a data acquisition method that the authors used in a project on modeling driver behavior for microscopic traffic simulations. An advanced instrumented vehicle was employed to collect driver-behavior data, mainly car-following and lane-changing patterns, on Swedish roads. To eliminate the measurement noise in acquired car-following patterns, the Kalman smoothing algorithm was applied to the state-space model of the physical states (acceleration, speed, and position) of both instrumented and tracked vehicles. The denoised driving patterns were used in the analysis of driver properties in the car-following stage. For further modeling of car-following behavior, we developed and implemented a consolidated fuzzy clustering algorithm to classify different car-following regimes from the preprocessed data. The algorithm considers time continuity of collected driver-behavior patterns and can be more reliably applied in the classification of continuous car-following regimes when the classical fuzzy C-means algorithm gives unclear results.

Driver behavior plays an important role in modeling vehicle dynamics in a traffic simulation environment. To study one element of the general driver behavior, that of car following, an advanced instrumented vehicle has been applied in dynamic data collection in real traffic flow on Swedishroads. This paper briefly introduces our car following data collection and smoothing methods. Moreover, we introduce spectrum analysis methods based on Fourier analysis of car following data to estimate driver reaction times, a crucial parameter of driver behavior. As an example, we calibrate a generalized GM-type model, an extension of the classical nonlinear GM model, in stable following regime based on the estimated driver reaction times. The calibrated model is then evaluated by closed-loop simulations.

An instrumented vehicle has been used to study car-following behavior on Swedish motorways. In this study, the previous data collection and pre-processing work were briefly reviewed. To understand the driving behavior in the car-following stage more clearly, the collected time series were classified into a number of regimes using unsupervised fuzzy clustering methods. Then, the statistical relations between the driver acceleration response and the perceptual variables in each regime were analyzed using correlation and regression methods. It was found that regime classification helps discern the behavioral variance between those regime clusters. According to the data analysis, some of the car-following regimes, for example, opening and braking, can be described adequately in the statistical sense by a linear regression model (Helly's model). Therefore, a multiple regime car-following model with simple model forms, for example, linear models, has the potential to robustly represent the general car-following behavior in most regimes.

Intelligent speed adaption (ISA) is one type of vehicle-based intelligent transportation systems (ITS), which warns and regulates driving speed according to the speed limits of the roads. Early field studies showed that ISA could reduce general mean speed levels and their variances in different road environments. This paper studies the effects of various ISA penetration grades on pedestrian safety in a single lane road. A microscopic traffic simulation tool, TPMA, was further developed and used to implement different ISA penetration grades. Momentary spot speed and traffic flow data are first logged in the traffic simulation for later prediction of pedestrian safety. Then a hypothetical vehicle-pedestrian collision model is extended from early researches in order to estimate two safety indicators: probability of collision, and risk of death. Finally, Monte Carlo method is applied iteratively to compute those safety indices. The computational result shows that raising ISA penetration in traffic flow will reduce both the probability of mid-block collision between vehicle and pedestrian and the risk of death in the collision accidents. Furthermore, the decrease of the risk of death will be more prominent than that of the collision probability according to this method.