A Study of Driver Behavior Inference Model at Time of Lane Change using Bayesian Networks

Recent years have brought hope that driving support systems tailored to the characteristics of each driver can be developed. To accomplish this, a driver model must be constructed that considers the driver's psychological function when inferring driver behavior. This paper thus proposes a method to infer driver behavior by capturing time-series steering angle data at the time of lane change. The proposed method uses a static type conditional Gaussian model on Bayesian networks. By using this method, if the driver behavior of the subject and learned data nearness of features (norms) are below a certain level, it is possible to infer driver behavior with nearly 100% probability. Moreover, compared to the HMM models, this method reduces the rate of incorrect inference inclusion.