Discovery of Driving Behavior Patterns Based on Multivariate Time Series

Zhenlong Li,Xiaohua Zhao,Jian Rong,Jianming Ma

2014-01-01

Abstract:This paper addresses the problem of driving behavior clustering and pattern discovery based on multivariate time series. Its aim is to discover and extract the patterns that reflect driving behavior from the vehicle movement measures. First, vehicle movement measures (lateral position and vehicle speed) were collected using a driving simulator. Vehicles’ lateral position and speed were recorded at each time point sequentially over time. A multivariate time series (MTS) was naturally formed. The multivariate time series data were obtained from a study that involved 22 subjects in a driving simulator located in the Transportation Research Center, Beijing University of Technology. Second, the time series was segmented based on series important points (SIP). Naturally, the slope can be calculated from the two points of the line segment. Third, lateral movements of vehicles were clustered into five patterns, which are “fast lateral movement to the right,” “slow lateral movement to the right,” “relatively unbiased movement or relatively forward movement,” “slow lateral movement to the left,” and “fast lateral movement to the left.”, using K-means clustering according to the slope of each segment of lateral position. Longitudinal movements were also clustered into five patterns according to the slope of each segment of vehicle speed. Lateral movements and longitudinal movements are interaction and co-movement. Thus, each pattern of lateral movements was cross-combined with each pattern of longitudinal movements. Twenty-five different patterns were discovered and extracted. The experimental results demonstrate that discovery of driving behavior patterns based on multivariate time series is feasible and effective.