Application of Real Field Connected Vehicle Data for Aggressive Driving Identification on Horizontal Curves

Abstract

The emerging technology of connected vehicles generates a vast amount of data that could be used to enhance roadway safety. In this study, we focused on safety applications of a real field connected vehicle data on a horizontal curve. The database contains connected vehicle data with instrumented vehicles that were carried out on public roads in Ann Arbor, Michigan. Horizontal curve negotiations are associated with a great number of accidents, which are mainly attributed to driving errors. Aggressive/risky driving is a contributing factor to the high rate of crashes on horizontal curves. Using basic safety message (BSM) data in connected vehicle dataset, this study modeled aggressive/risky driving while negotiating a horizontal curve. The model was developed using the machine learning method of Random Forest to classify the value of time to lane crossing (TLC), a proxy for aggressive/risky driving, based on a set of motion-related metrics as features. Three scenarios were investigated considering different TLCs value for tagging aggressive driving moments. The model contributed to high detection accuracy in all three scenarios. This suggests that the motion-related variables used in the random forest model can accurately reflect drivers' instantaneous decisions and identify their aggressive driving behavior. The results of this study inform the design of warning/feedback systems and control assistance from unsafe events which are transmittable through vehicles-to-vehicles (V2V) and vehicles-to-infrastructure (V2I) applications.

Review of Usage of Real-World Connected Vehicle Data

“…Three recent studies used the CV data from the SPMD program to propose methodologies that use critical information from the instantaneous BSM exchanges between CVs and roadside equipment to determine repeatable driving behaviors ( 14–16 ). Liu and Khattak ( 14 ) investigated the longitudinal and lateral motion of the driving decision from BSMs and established reasonable thresholds to identify potentially dangerous events such as hard accelerations or braking, and quick lane changes.…”

“…Machine learning methods can identify the importance of motion-related variables in classifying driving data into aggressive and normal driving patterns ( 17, 18 ). Jahangiri et al ( 16 ) applied machine learning to the CV data from the SPMD program to identify aggressive driving patterns on horizontal curves. They used the random forest method of machine learning to develop an aggressive driving detection model based on a time-to-lane crossing metric, under three scenarios.…”

“…Khattak and colleagues ( 14, 15 ) used the one-day sample datasets but deleted some observations that contained errors. Jahangiri et al ( 16 ) used a one-month dataset. Larger datasets require more processing capacity to produce results in a reasonable amount of time, but they can facilitate the removal of outlier situations, such as extremely bad weather conditions that could bias the results.…”

“…Data Elements of the DataFrontTargets File Indicates whether an obstacle in the vehicle's path is the closest. information from the instantaneous BSM exchanges between CVs and roadside equipment to determine repeatable driving behaviors(14)(15)(16).Liu and Khattak (14) investigated the longitudinal and lateral motion of the driving decision from BSMs and established reasonable thresholds to identify potentially dangerous events such as hard accelerations or braking, and quick lane changes. They used the DAS datasets and the 10-Hertz motion data that contain speed along with longitudinal and lateral acceleration to visualize the driving behavior.…”

“…Khattak and Wali(15) Markov switching dynamic regression model (time-series analysis): Quantification and prediction of driving patterns. Jahangiri et al(16) Random forest classification of risky driving behaviors. Surrogate safety measures He et al(20) Negative binomial regression model: Statistical relationship between the link developed safety surrogate measures and crash frequency.…”