Embedded Files
How can we effectively transform big data from traffic apps, traffic video, and other traffic sensory data to improve driving conditions?
How can we effectively transform big data from traffic apps, traffic video, and other traffic sensory data to improve driving conditions?
Project 1: Deep Learning Software for Traffic State Prediction
Project 1: Deep Learning Software for Traffic State Prediction
The Problem
Traffic congestion causes Americans to lose billions of hours per year and poses a threat to economic prosperity due to the waste of time and fuel. Current strategies for estimating traffic states use point-based sensors and data from GPS navigation. Specifically, traffic signal systems are one area where improvement could yield gains on traffic congestion. Recent research has shown that additional data could help to improve traffic signal coordination.
Did you know? It is estimated that Americans lose up to 4.2 billion hours in traffic annually (Texas Transportation Institute, 2008)
Did you know? It is estimated that Americans lose up to 4.2 billion hours in traffic annually (Texas Transportation Institute, 2008)
Our Research
We have identified deep learning "deep neural networks" (DNNs) as a potential tool for improving traffic signal systems and traffic estimation. DNNs would allow for better use of existing traffic video data and other data to not only improve signal systems and estimation, but also learn from the data and improve automatically over time. We will train the DNNs using data collected from traffic videos, Bluetooth sensors, and loop detectors and also develop data fusion methods to improve the data quality for a video analytics pipeline.
Our Goals
This project aims to develop an edge-computing, deep-learning software using video, loop detector, and Bluetooth sensor data to better estimate traffic on arterial roads. It will leverage the best of DNN and machine learning models to produce a video analytics pipeline for video and image analysis. We anticipate producing a prototype application of an advanced traffic signal controller using edge computing design.
A final report to the NCDOT will detail the results of the research.
How does this research benefit North Carolina?
The results of this research will equip the NCDOT to better manage the more than 380 coordinated traffic signal systems they manage. Better coordinated traffic signals can save all drivers many hours on the road.
Page updated
Report abuse