Studying the Dynamic Sight Distance Problem with a Machine Learning Algorithm
The Dynamic Sight Distance (DSD) problem involves the dynamically allocated sight distance as a left-turning vehicle makes its move to clear a signalized intersection during the permissive phase whose line-of-sight is obstructed from the presence of stationary vehicles in opposing left-turning lanes. In this paper the authors study the DSD problem with a Machine Learning (ML) algorithm called Random Forest (RF) classifier. In their previous works the authors used limited field data to measure gap sizes and a driver’s willingness to accept or reject critical gaps to formulate the DSD problem; and calculated the probability of line-of-sight in the analysis. Field data from 10 signalized intersections from Maryland with unprotected left turns were used in the study. This paper extends the previous methodology by introducing additional variables attributed to gap acceptance, such as obstruction angle, driver’s age, number of queued vehicles, and presence or absence of a peak hour. The results show that presence or absence of peak hour and number of queued vehicles are highly correlated to the driver’s decision-making in accepting or rejecting a gap. Several simulation runs are conducted to improve the accuracy of the model. The authors conclude that the RF classifier is highly effective and better than traditional Raff’s method in predicting a driver’s behavior to accept or reject a gap associated with the DSD problem. Future works may include extending the methodology for arterial network with many signalized intersections with DSD issues, traffic signal optimization, simulation related studies, Vehicle to Vehicle communication, and Vehicle Infrastructure Integration.
- Record URL:
-
Supplemental Notes:
- This paper was sponsored by TRB committee AKD10 Standing Committee on Performance Effects of Geometric Design.
-
Corporate Authors:
Transportation Research Board
, -
Authors:
- Jha, Manoj K
- Ogallo, Hellon
-
Conference:
- Transportation Research Board 100th Annual Meeting
- Location: Washington DC, United States
- Date: 2021-1-5 to 2021-1-29
- Date: 2021
Language
- English
Media Info
- Media Type: Digital/other
- Features: Figures; References; Tables;
- Pagination: 16p
Subject/Index Terms
- TRT Terms: Algorithms; Gap acceptance; Highway safety; Left turns; Line of sight; Machine learning; Permissive phasing; Sight distance; Signalized intersections
- Geographic Terms: Maryland
- Subject Areas: Design; Highways; Safety and Human Factors;
Filing Info
- Accession Number: 01764431
- Record Type: Publication
- Report/Paper Numbers: TRBAM-21-03783
- Files: TRIS, TRB, ATRI
- Created Date: Feb 4 2021 4:48PM