Modeling aggressive driving behavior based on graph construction

The occurrence of aggressive driving behavior is a random process among time-varying transversion. For modeling aggressive driving behavior, previous studies have typically relied on summative statistics (e.g. means and S.D.) as the inputs which fall short in extracting features of driving behaviors that are in a time-series pattern. The raw data are used to extract the pieces of the graph. Each graph represents a specific driving trip that includes driver characteristics, environment information, and driving behavior variables. Graphs of trips were constructed based on the Shanghai Naturalistic Driving Study data. 17 variables related to aggressive driving are extracted and converted into elements (pieces as described in the paper) in the construction graphs. Regression models are used for verifying the performance of using graph construction as the input in modeling. The result shows that a 5-sec time window is suitable for aggressive driving behavior modeling. 11 variables (speed, longitudinal acceleration, lateral acceleration, lateral displacement, gender, age, distracted, drowsy, weather, horizontal alignment, time-to-collision) are selected for graph construction based on their significance. Both summative statistics and graph construction are applied. The result shows that graph construction achieves better performance over models based on summative statistic inputs (mean plus standard deviation, and mean only). Based on weights of variables, time-to-collision is found to be the factor which affects aggressive driving most. This method can be used in real-world applications to improve driving safety by alerting drivers of aggressive behavior identified in real time with Advanced Driver Assistance System applications.

Language

  • English

Media Info

Subject/Index Terms

Filing Info

  • Accession Number: 01844215
  • Record Type: Publication
  • Files: TRIS
  • Created Date: Apr 26 2022 5:07PM