Predicting Driver Behavior at a Crosswalk Using Kinematic Time-Series Data

• Activity Number: 131 - The Future of Transportation: The Predicting Power of Driver Behavior Data
• Type: Topic Contributed
• Date/Time: Monday, August 3, 2020 : 1:00 PM to 2:50 PM
• Sponsor: Transportation Statistics Interest Group
• Author(s): Huizhong Guo* and Linda Ng Boyle
• Companies: University of Washington and University of Washington
• Keywords: Naturalistic driving data; Driver behavior; Prediction; Crosswalk

Abstract

Crosswalks provide the right-of-way to pedestrians to increase their safety on the road. However, the likelihood of a vehicle-pedestrian crash depends on the ability to see the pedestrian. Pedestrians could be overlooked if drivers are traveling too fast or if the pedestrian dashes onto the crosswalk unexpectedly. In this study, we seek to use drivers’ kinematic time-series data to predict whether they would slow down before the crosswalk or not. More specifically, we compare the performance of a feature-based method (e.g., SVM) to a model-based method (e.g., HMM). We examine how many meters away from the crosswalk can be reasonably predicted given a driver’s speed selection. This model can be used in transportation systems to provide timely feedback to drivers at signalized crosswalks. It can also provide additional benefits to pedestrians to warn them of high-speed approaching vehicles and helps them stay safe.