Model
Digital Document
Publisher
Florida Atlantic University
Description
The United States has been going through a road accident crisis for many
years. The National Safety Council estimates 40,000 people were killed and 4.57
million injured on U.S. roads in 2017. Direct and indirect loss from tra c congestion
only is more than $140 billion every year. Vehicular Ad-hoc Networks (VANETs) are
envisioned as the future of Intelligent Transportation Systems (ITSs). They have a
great potential to enable all kinds of applications that will enhance road safety and
transportation efficiency. In this dissertation, we have aggregated seven years of real-life tra c and
incidents data, obtained from the Florida Department of Transportation District 4.
We have studied and investigated the causes of road incidents by applying machine
learning approaches to this aggregated big dataset. A scalable, reliable, and automatic
system for predicting road incidents is an integral part of any e ective ITS. For this
purpose, we propose a cloud-based system for VANET that aims at preventing or at
least decreasing tra c congestions as well as crashes in real-time. We have created,
tested, and validated a VANET traffic dataset by applying the connected vehicle
behavioral changes to our aggregated dataset. To achieve the scalability, speed, and fault-tolerance in our developed system, we built our system in a lambda architecture
fashion using Apache Spark and Spark Streaming with Kafka.
We used our system in creating optimal and safe trajectories for autonomous
vehicles based on the user preferences. We extended the use of our developed system in
predicting the clearance time on the highway in real-time, as an important component
of the traffic incident management system. We implemented the time series analysis
and forecasting in our real-time system as a component for predicting traffic
flow.
Our system can be applied to use dedicated short communication (DSRC), cellular,
or hybrid communication schema to receive streaming data and send back the safety
messages.
The performance of the proposed system has been extensively tested on the
FAUs High Performance Computing Cluster (HPCC), as well as on a single node
virtual machine. Results and findings confirm the applicability of the proposed system
in predicting traffic incidents with low processing latency.
years. The National Safety Council estimates 40,000 people were killed and 4.57
million injured on U.S. roads in 2017. Direct and indirect loss from tra c congestion
only is more than $140 billion every year. Vehicular Ad-hoc Networks (VANETs) are
envisioned as the future of Intelligent Transportation Systems (ITSs). They have a
great potential to enable all kinds of applications that will enhance road safety and
transportation efficiency. In this dissertation, we have aggregated seven years of real-life tra c and
incidents data, obtained from the Florida Department of Transportation District 4.
We have studied and investigated the causes of road incidents by applying machine
learning approaches to this aggregated big dataset. A scalable, reliable, and automatic
system for predicting road incidents is an integral part of any e ective ITS. For this
purpose, we propose a cloud-based system for VANET that aims at preventing or at
least decreasing tra c congestions as well as crashes in real-time. We have created,
tested, and validated a VANET traffic dataset by applying the connected vehicle
behavioral changes to our aggregated dataset. To achieve the scalability, speed, and fault-tolerance in our developed system, we built our system in a lambda architecture
fashion using Apache Spark and Spark Streaming with Kafka.
We used our system in creating optimal and safe trajectories for autonomous
vehicles based on the user preferences. We extended the use of our developed system in
predicting the clearance time on the highway in real-time, as an important component
of the traffic incident management system. We implemented the time series analysis
and forecasting in our real-time system as a component for predicting traffic
flow.
Our system can be applied to use dedicated short communication (DSRC), cellular,
or hybrid communication schema to receive streaming data and send back the safety
messages.
The performance of the proposed system has been extensively tested on the
FAUs High Performance Computing Cluster (HPCC), as well as on a single node
virtual machine. Results and findings confirm the applicability of the proposed system
in predicting traffic incidents with low processing latency.
Member of