Date of Award

8-2022

Document Type

Dissertation

Degree Name

Ph.D.

Department

Systems Engineering

Committee Chair

Min-Wook Kang, Ph.D.

Abstract

Drivers’ indecisions within the dilemma zone (DZ) during the yellow interval is a major safety concern of a roadway network. The present study develops a systematic framework of a machine learning (ML) based dynamic dilemma zone protection (DZP) system to protect drivers from potential intersection crashes due to such indecisions. For this, the present study first develops effective methods of quantifying DZ using important site-specific characteristics of signalized intersections. By this method, high-risk intersections in terms of DZ crashes could be identified using readily available intersection site-specific characteristics. Afterward, the present study develops an innovative framework for predicting driver behavior under varying DZ conditions using ML methods. The framework utilizes multiple ML techniques to process vehicle attribute data (e.g., speed, location, and time-of-arrival) collected at the onset of the yellow indication, and eventually predict drivers’ stop-or-go decisions based on the data. The DZP system discussed in the present study has two major components that work with synergy to ensure the total safety of a DZ affected vehicle: dynamic green extension (DGE), and dynamic green protection (DRP) system. Based on the continuous vehicle tracking data, the DGE system uninterruptedly monitors vehicle within the DZ and xiv predict vehicles that may face the decision dilemma if there is a sudden transition from green signal to yellow. After detecting such vehicles, the DGE system provides an exact amount of extended green time so that the detected vehicles could safely clear the intersection without any hesitation. There could be some vehicles that may end up running the red light due to various limitations. In this case, the DRP system provides an extended amount of all-red extensions after predicting potential red light running vehicles to nullify the likelihood of any intersection crashes. After the development, the DZP system is then implemented in several selected intersections in Alabama. Performance assessments are accomplished for the to see the safety and operation impact of the DZP system in implemented sites. The comprehensive assessment of the DGE system is accomplished with ten performance measures, which include percent green arrivals, percent yellow arrivals, percent red arrivals, dilemma zone length, and red-light running vehicles before and after the system implementation. Results show that the DGE system could significantly improve the overall intersection safety and efficiency. A short-term study on performance assessment of DRP systems shows that such a driver behavior prediction method could effectively predict 100% red-light-runners as well as efficiently provide the required amount of clearance time without hampering overall intersection efficiency. Based on the outcomes from the performance assessments of the DGE and DRP systems, it is safe to say the machine learning based DZP system would be able to promote intersection safety by protecting the dilemma zone impacted vehicles from potential intersection crashes as well as enhance the operational performance of intersections by intelligently allocate exact right-of-way to the vehicles and reducing the overall delays.

Share

COinS