Prediction of Non-Motorized Vehicle’s Overtaking Trajectory on Shared Bicycle Lane

Current research on non-motorized vehicles is not as extensive as motor vehicles, particularly in regard to the prediction of non-motorized vehicle trajectory. The accurate prediction of non-motorized vehicle trajectory is essential for many roadway safety applications such as the autonomous vehicle’s path planning in road space shared with non-motorized vehicles and developing non-motorized vehicle related crash avoidance systems. Specifically, this study aims to develop models to predict a non-motorized vehicle’s overtaking trajectory. Trajectory data of 1179 non-motorized vehicles from 168 overtaking events were collected to develop the model. A long short term memory (LSTM) model was established based on variables including the average distance, average speed difference and other indicators that reflect the relationships between non-motorized vehicles. Its results were compared with a Random Forest model which was used as a base model. The results demonstrated that the LSTM neural network obtained more accurate results with a 98% of accuracy for the lateral position, compared with a 91% of accuracy of the Random Forest model. Further segmentation of trajectories and independent modeling of trajectories after overtaking begins obtained more accurate results on future location prediction, and the prediction accuracy of the lateral coordinates after 0.5s was increased from 95% to 97%. Using two-second historical information, the mean absolute error of the trajectory position prediction after 1 / 12s and 0.5s was respectively 0.208m and 0.429m. The findings of this study showed the promising of the LSTM on the prediction of non18 motorized vehicle trajectory.

Language

  • English

Media Info

  • Media Type: Digital/other
  • Features: Figures; Photos; References; Tables;
  • Pagination: 19p

Subject/Index Terms

Filing Info

  • Accession Number: 01764285
  • Record Type: Publication
  • Report/Paper Numbers: TRBAM-21-03019
  • Files: TRIS, TRB, ATRI
  • Created Date: Dec 23 2020 11:24AM