Segmentation and Clustering of Car-Following Behavior: Recognition of Driving Patterns

Abstract

Driving behavior can be influenced by many factors that are not feasible to collect in driving behavior studies. The research presented in this paper investigates the characteristics of a wide range of driving behaviors linking driving states to the drivers' actions. The proposed methodology is structured such that a known state can be linked to multiple actions, thus accounting for the effects of unknown driving state factors. A two-step algorithm was developed and used for the segmentation and clustering of driving behaviors. The algorithm segments and clusters car-following behaviors based on eight state-action variables: the longitudinal acceleration, the lateral acceleration, the yaw rate, the vehicle speed, the lane offset, the yaw angle, the range, and the range rate. The results of this methodology are state-action clusters that define the driving pattern of drivers. The sample used in this paper included 20 different drivers, i.e., ten car and ten truck drivers. The results revealed that behavior patterns were different between car drivers but not between truck drivers. The results also show that car drivers exhibit behaviors that are unique to each driver, whereas truck drivers show a common driving pattern. The characteristics and frequency of recognized driving patterns are provided in this paper, along with the corresponding modeling parameters of each pattern.Index Terms-Clustering methods, data mining, driver behavior, traffic flow theory.

A Novel Model-Based Driving Behavior Recognition System Using Motion Sensors

“…That is the reason why this work can only recognize seven different sub-typical kinds of driving events without having a better performance in terms of accuracy. B. Higgs and M. Abbas defined the driver behavior as a map function with current traffic state as the argument and driver action as the dependent variable [19]. This work utilized segmentation and clustering to decompose the single map function in traditional car-following models into several different functions.…”

Car-following Behavior Model Learning Using Timed Automata

“…In the near future, we will implement the segmentation and clustering approach of Higgs and Abbas (2015) and make a comparison with ours. Our model can be used for subject drivers' decision making by recognizing or predicting surrounding vehicles' car-following states.…”

“…Our work is motivated by Higgs and Abbas (2015). In their paper, they first segmented the time series data by means of change point detection, then the mean values representing the segmented piece-wise data were clustered using k-means.…”

A Channel Allocation Mechanism for Cellular Networks

“…The vehicle movement and MS communication traces are generated by a traffic simulation program VISSIM. This study considered a highway scenario that is characterized by the Wiedemann "psycho-physical" car-following model and lane changing model [34][35][36]. The conceptual development and limited available data can be considered by the Wiedemann model to generate traffic steam data for highways and freeways [27,28].…”