Abstract
Lane changing is considered as one of the most dangerous driving behaviors because drivers have to deal with the traffic conflicts on both the current and target lanes. This study aimed to propose a method of predicting the driving risks during the lane-changing process using drivers’ physiology measurement data and vehicle dynamic data. All the data used in the proposed model were obtained by portable sensors with the capability of recording data in the actual driving process. A hidden Markov model (HMM) was proposed to link driving risk with drivers’ physiology information and vehicle dynamic data. The two-factor indicators were established to evaluate the performances of eye movement, heart rate variability, and vehicle dynamic parameters on driving risk. The standard deviation of normal to normal R–R intervals of the heart rate (SDNN), fixation duration, saccade range, and average speed were then selected as the input of the HMM. The HMM was trained and tested using field-observed data collected in Xi’an City. The proposed model using the data from the physiology measurement sensor can identify dangerous driving state from normal driving state and predict the transition probability between these two states. The results match the perceptions of the tested drivers with an accuracy rate of 90.67%. The proposed model can be used to develop proactive crash prevention strategies.
Highlights
The drivers’ physiological data and vehicle dynamic data were collected as the input dataset of the driving risk prediction model
The initial settings the hidden Markov model (HMM) are listed as follows: the number of hidden states N was two; the number of observable parameters M was 4; the initial state transition probability matrix A0 could be obtained by the averaging method shown in Equation (4); the observed probability matrix Bk could be calculated from the kth observation parameters; the initial probability distribution of the lth driver Πl could be obtained by the averaging method in Equation (5)
The initial settings of the HMM are listed as follows: the number of hidden states N was two; the number of observable parameters M was 4; the initial state transition probability matrix A0 could be obtained by the averaging method shown in Equation (4); the observed probability matrix Bk could be calculated from the kth observation parameters; the initial probability distribution of the lth driver Πl could be obtained by the averaging method in Equation (5)
Summary
During the lane-changing process, drivers have to deal with complex traffic conflicts between current and target lanes, leading to increased driving workload [1]. Sensors 2019, 19, 2670 driving behaviors [7] These models usually use the driver’s physiological information to represent the driver’s current natural status, rather than in the prediction of driving risk in the near future. The principal reason behind not using the driver’s physiological indicators in the driving risk prediction models might be that there is no direct connection between these indicators and the driving risk In this way, we need to develop a model that can connect the driver’s physiology status with the driving risk for a driving task with a heavy workload, such as the lane-changing process. The proposed model should take full consideration of the complex interactive effects among driving behaviors, driver’s physiology status, vehicle dynamics, and surrounding traffic dynamics [8], whose measurement data are collected by multiple sensors in real time
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
