Date of Award
Master of Science
Electrical and Computer Engr
Traffic situations are continuous, uncertain, highly dynamic and partially observable, and they affect the day-to-day lives of people in a society. A worthwhile endeavor is to develop algorithms that can predict abnormal traffic situations by exploiting data from the myriad of sensors on the streets, in vehicles and in smartphones, leading to smoother flow of traffic. Unfortunately, the large volumes of microscopic (i.e. individual vehicle-level) data required for developing statistical/machine learning algorithms cannot be collected from the field by the public. The data collected by transportation agencies is either macroscopic or not widely available. In this thesis, a framework is developed for simulating large-scale traffic data using a microscopic simulation model and limited real-world data. Five kinds of sensors are simulated: inductor loop detector, lane area detector, multi-entry multi-exit detector, Bluetooth, and edgebased traffic measure. Data is simulated using this framework from multiple sensors over an area covering Montgomery County and Prince George County in Washington DC for 720 hours (30 days). The synthesized data is validated with respect to real-world data for volume and speed. Widely-used classifiers are used to recognize eight traffic events, namely Collision, Disabled Vehicle, Emergency Roadwork, Injuries Involved, Obstructions, Road Maintenance Operations, Traffic Signal Not Working and with no events in the synthesized dataset with high accuracy. Given limited real-world microscopic traffic data from a particular area, this framework is the first of its kind that can simulate data from multiple kinds of sensors over a very long duration with high-fidelity to the given data.
dissertation or thesis originally submitted to the local University of Memphis Electronic Theses & dissertation (ETD) Repository.
Khirwadkar, Vrinda, "Simulating Large-Scale Microscopic Traffic Data" (2020). Electronic Theses and Dissertations. 2064.