TIMELI: Traffic Incident Management Enabled by Large-data Innovations
- Neal Hawkins | 515-294-7733 | firstname.lastname@example.org
- Srikanta Tirthapura | email@example.com
- Soumik Sarkar | firstname.lastname@example.org
- Stephen Gilbert | email@example.com
Start date: 09/01/16
End date: 08/31/19
Sponsor(s): National Science Foundation
Partner(s): TransCore and Iowa DOT
About the research
This National Science Foundation Partnerships for Innovation: Building Innovation Capacity (PFI:BIC) project is titled TIMELI: Traffic Incident Management Enabled by Large-data Innovations. The goal and outcome of TIMELI is to use emerging large scale data analytics techniques reduce the number of road incidents through proactive traffic control and to minimize the impact of individual incidents that do occur through early detection, response, and traffic management and control.
The objectives are to develop TIMELI and to integrate it into an existing traffic incident management (TIM) system. These objectives will be accomplished by using the following methods:
(1) Define TIM user requirements and identify bottlenecks in technician tasks using human factors research
(2) Develop a prototype that includes a big-data-enabled back-end solution, an analytics engine, and a front-end interface
(3) Conduct testing, evaluation, and integration within the Iowa Department of Transportation's (DOT's) existing TIM environment.
The areas of expertise for TIMELI are big data analysis and processing, transportation modeling, traffic management, machine learning, anomaly detection, human cognition and training, and interface design and visualization. The test bed will be the Center for Transportation Research and Education's (CTRE's) fully functional traffic operations laboratory that is connected to the Iowa DOT’s data streams.
TIMELI’s multiple innovations will transform current TIM systems by creating a smart and reliable decision-assist system used to monitor traffic conditions in real time, proactively control risk using advisory control, quickly detect traffic incidents, identify the location and potential cause of these incidents, suggest traffic control alternatives, and minimize cognitive bottlenecks for TIM operators. This will be achieved using end-to-end machine learning for situational awareness, the design and rapid solution of geo-temporally aware traffic models using partial differential equations, stochastic model predictive control, and user-centric advanced visualization techniques for decision assistance.
Current technology gaps in data handling and archiving, analysis for decision support, and the design of output formats will be addressed using big data technologies. Multiple large data streams will be ingested and data analytics will be performed for quality assurance and anomaly detection. New algorithmic approaches, machine learning, and a stochastic framework will be used to detect anomalous outliers and implement context-sensitive traffic models. An advanced human machine interface will provide information visualization and decisions recommendations in an intuitive format to minimize any cognitive bottlenecks.
A quarter of the congestion on US roads is due to traffic incidents often caused by a crash, an overturned truck, or stalled vehicles. Congestion costs the commercial trucking industry $9.2 billion annually. Incidents have also been shown to increase the risk of secondary crashes by 2.8 percent with every passing minute of induced congestion. To address these economic and safety issues, TIM centers monitor roadways for traffic incidents, coordinate incident response, and provide traffic management and control to minimize the impacts of traffic incidents; yet, these systems are currently limited. More effective data-driven TIM, with user-centric information visualization and improved analytics and machine learning, can reduce the duration and impacts of incidents and improve the safety of motorists, crash victims, and emergency responders. This can also reduce the TIM technician fatigue and thereby reduce their turnover rates.
Real-world datasets generated during the project will be used in developing education material for classes currently taught by the team. The research results will be disseminated through multiple channels, including technical presentations and papers.