Bayesian Multinomial Latent Variable Model to Detect Driver Distraction at Intersections

• Activity Number: 476 - Distracted Driving and Other Transportation Considerations
• Type: Contributed
• Date/Time: Wednesday, July 31, 2019 : 8:30 AM to 10:20 AM
• Sponsor: Transportation Statistics Interest Group
• Author(s): Ning Li* and Linda Ng Boyle
• Companies: University of Washington and University of Washington
• Keywords: Bayesian; Latent variable modeling; Naturalistic driving data; Driver distraction; Driver monitoring

Abstract

Driver distraction at intersections can contribute to many vehicular crashes and fatalities. The ability to classify the driver distraction state would be useful in understanding how drivers behave at various intersections, and provide guidance for enhancing safe driving in these areas. In this study, an algorithm is developed for predicting different driver distraction activities at various intersections that have different traffic control. The data comes from a naturalistic driving study collected from Ann Arbor, Michigan. We propose a multinomial Bayesian latent variable model that is able to consider the latent contributions of the roadway and environmental characteristics, and is then able to predict the probabilities for different distraction behaviors (e.g., cell phone operation, talking, eating). The estimation of the model parameters is based on a Markov Chain Monte Carlo (MCMC) algorithm. Latent variables allow the prediction of driver distraction to be significantly improved, which provides us a better understanding of the driver’s response to varying traffic control devices.