Machine learning application powering automation of efficient traffic operation
Date
2022-12
Authors
Zarindast, Atousa
Major Professor
Advisor
Sharma, Anuj
Day, Christopher
Wood, Johnathon
Dong, Jing
Sarkar, Soumik
Committee Member
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Developing intelligent transportation systems attracted a lot of attention from
practitioners to researchers as it provides decision support systems for efficient, automated and
smart transportation management. The idea of Intelligent transportation systems originally was
born in 1980s by a small group of professionals to impact computing and communications
techniques [1]. Term “Big Data” originally means data that cannot be stored, processed and
analyzed with traditional methods. Big data analytics are processes that collects, manages,
analyze and visualize continuously evolving data [2]. Big data in transportation systems enabled
massive development and deployment of sensors, connected and autonomous vehicles and other
resources. In order to manage big data transportation systems in intelligent transportation
systems data driven models are needed, therefore big data algorithm is developed and deployed
in traffic management that changed the intelligent transportation management systems. In this
study we provide intelligent transportation systems using machine learning in era of big data for
three different transportation sensors and sensor data type. The first section of this study focuses
on providing a data driven method for congestion identification using probe data. A big data
method for a well-known problem without a definition in literature. A data driven algorithm for
congestion identification is developed that relies on data itself. In the second section a data
driven method to reverse engineer the detector meta data is proposed. Measuring traffic
intersections performance measure relies on detector information data. This study provides a
data-driven method for automatic identification of detector types and phase assignments at
signalized intersections using high-resolution event data. This study proposes a machine learning
method, Occupancy Pattern Association (OPA), to identify the type of detector (stop bar or
viii
upstream) and the assignment of phases actuated by the detector. The effectiveness of the
proposed algorithms is illustrated by processing a wide range of intersections across three
different corridors in different states of Utah, Nebraska and Iowa. The accuracy of detector type
identification is 88 % or higher among all intersections, and detector configurations at seven
intersections are identified with 100% accuracy. The proposed method holds promise for
addressing the challenge of developing and maintaining an inventory of detector configurations.
In the last section of current study, we consider camera sensors and propose a method for
vehicle detection and tracking in different scenarios and camera settings. This study proposes a
tracking system for the transportation field utilizing state-of-the-art detection heads and tracking
algorithms. We utilize detection heads (YOLO5, FRCNN, DETR) with tracking algorithms-
(SORT, ByteTrack) under different camera settings and three weather scenarios. The tracking
results are summarized regarding Recall and precision, IDF1, IDP, IDR, MOTA, MOTP,
tracking consistency, and processing time. In this study, we utilize novel Transformer
architecture, a deep learning network based on the attention mechanism.
Series Number
Journal Issue
Is Version Of
Versions
Series
Academic or Administrative Unit
Type
dissertation