Abstract
Low user acceptance is one of the fundamental problems for popularizing advanced driver assistance systems (ADAS). Systems that are developed for the majority of drivers have to possess stationary characteristics and be conservative for safety reasons. However, the drivers with disparate driving styles possess different risk cognition of lane change behavior; therefore, such systems with stationary characteristics may cause frequent interference to aggressive drivers or may be perceived as a radical system by conservative drivers. An ADAS that adapts to the characteristics of individual drivers during lane change maneuvers will be more effective and more acceptable to drivers. In this study, we developed an adaptive algorithm that learns the characteristics of individual drivers during lane changes and determines the optimal threshold online to adapt to different drivers. Signal detection theory (SDT) was employed and the results of the accuracy, false negative rate, and false positive rate were used to capture the drivers' lane change behavior characteristics. A learning stage and a threshold fluctuation stage were designed in the adaptive algorithm to determine the optimal warning threshold and amended the optimal warning threshold based on changes in the drivers' behaviors. We evaluated the proposed algorithm by conducting the actual vehicle tests with a total of three participants. The offline statistical analysis results of the participants' lane change characteristics were compared with the online results of the warning threshold adjustments from the adaptive algorithm; the comparison results indicated that the adaptive algorithm could effectively capture the drivers' lane change characteristics and determine an appropriate warning threshold. The findings provide an improvement in the performance of the lane change warning (LCW) system and enhance people's acceptance of intelligent systems.
Highlights
Lane change warning (LCW) systems have been applied increasingly in intelligent vehicles to enhance the safety of lane changes and reduce driver workload [1]–[3]
In order to ensure that the proposed algorithm could accurately capture the driver’s lane change characteristics, after the actual vehicle experiments, we statistically analyzed the average TTC values of the participants who had completed lane change and yielded lane change in the learning stage and the fluctuation stage; we compared the offline statistical results with the optimal thresholds derived from the proposed adaptive algorithm based on the Signal detection theory (SDT) method
participant 1 (P1) possessed the highest value of Pl of almost 70% at a TTC threshold of 3.0 s, which indicated that P1 was the most conservative driver among the three participants; and the warning function was regarded as invalid on account of the safer warning threshold of P1 than the system’s initial value
Summary
Lane change warning (LCW) systems have been applied increasingly in intelligent vehicles to enhance the safety of lane changes and reduce driver workload [1]–[3]. Markov model (HMM), fuzzy model, support vector machine (SVM), Bayesian network, and Gaussian mixture model (GMM) [19], [20] These models are very effective in terms of model adaptability to different drivers and accuracy improvement due to sufficient training data based on naturalistic driving. Numerous self-learning algorithms of driving characteristics based on mathematical optimization methods have been developed to improve the performance and acceptance of longitudinal driver assistance systems (i.e., forward collision warning system) [36], [37], researches focused on the performance improvement of vehicle lateral safety systems (i.e., LCW system) are sparse. A self-learning method to determine a driver’s lane change characteristics based on signal detection theory (SDT) was proposed.
Sign-in/Register to view highlights & summary 
Published Version (
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