Emerging Technologies Featured Project: Rapidly deployable, low-cost traffic data and video collection device

Principal Investigator: Panos Michalopoulos, Department of Civil, Environmental, and Geo- Engineering
Co-Principal Investigator: Ted Morris, Minnesota Traffic Observatory

To meet the needs of transportation practitioners, planners, and researchers, a team from the University of Minnesota’s Department of Civil, Environmental, and Geo- Engineering and engineers  at the Minnesota Traffic Observatory developed a portable traffic data collection system that is easily deployable, non-intrusive, and inexpensive. This system is designed for temporary data gathering and video recording of vehicle movements at intersections and along arterial roadways, and for remote surveillance of areas of interest.

The system is useful in gathering a variety of data including traffic volume, speed, vehicle classification, turning movements, queue size, conflicting movements, and time headways. The video system also provides a visual record of traffic characteristics, accidents, and special situations. In addition, it has the potential of extracting turning movements automatically including optional lanes through advanced machine vision or radar sensors.

Because a single unit can cover an intersection of up to five lanes per incoming approach (20 incoming lanes total), the system is suitable for monitoring the vast majority of intersections.


Prototype development

The research team developed functional specifications for the portable data collection system by consulting a panel of experts in traffic data collection, including practitioners, researchers, and engineers.

The effects of camera mast height, camera position with respect to the intersection, and field of view were investigated by modifying an existing apparatus so that it could collect video recordings and deploying it at three urban intersection sites.

A prototype data recording apparatus was then developed, featuring a self-raising extensible mast and custom-fabricated base to elevate the camera to a maximum height of 28 feet above the road surface. The apparatus is designed to be secured by clamps to sign, light, or traffic signal poles.

An interface was developed to configure daily recording schedules and other hardware in order to utilize battery power as efficiently as possible. Approximately 40 hours of traffic video can be stored before battery swapping/recharging is necessary.

System deployment 1: Intersection of 4-lane divided highway and 2-lane arterial road.

System in operation, secured to signal pole on sidewalk.

System in operation, secured to signal pole on sidewalk.

Intersection diagram showing camera field of view and traffic movements.

Intersection diagram showingcamera field of view and traffic movements.

Camera view of intersection.

Camera view of intersection.

System deployment 2: Major suburban arterial intersection.

Prototype system in operation,

Prototype system in operation,secured to traffic signal pole.

Intersection diagram showing camera field of view and traffic movements.

Intersection diagram showing camerafield of view and traffic movements.

Camera view of intersection.

Camera view of intersection.


Testing

The system has been deployed and left unattended for periods varying from one to two weeks at a time at several selected typical urban intersections and one mid-block site to collect video under a variety of traffic and environmental conditions (this would be hyperlinked to figures). Standard traffic measurements (speeds, counts, vehicle classification) using a widely available commercial machine vision system were then extracted and compared with ground truth data.  A statistical analysis was then performed to determine the accuracy of the extracted measurements.

The findings of this study demonstrate volumes at arterials can be automatically extracted with an aggregate mean error within 10 percent of the ground truth measurements.  The mid-block experiment determined the ability of the apparatus to extract vehicle speed and class upstream of an intersection. The findings from that experiment indicated the lane closest to the apparatus matched ground truth speeds reasonably well, although accuracy degraded as the cross-lane distance from the camera increased.  Classification for private vehicles was correct more than 90% of the time, while larger vehicle class assignments resulted in much larger errors. Limitations and sources of errors affecting the measurement accuracy of the machine vision system were also studied. This project demonstrated that a rapidly deployable stand-alone, unattended apparatus for collecting traffic video data at or near arterial intersections is feasible and reliable.


Related Links


Acknowledgments

This research was supported by the University of Minnesota’s Intelligent Transportation Systems Institute with funding from United States Department of Transportation’s Research and Innovative Technologies Administration (RITA) University Transportation Centers (UTC) program. Ron Engh of Mn/DOT Signal Operations and Mike Spack of Spack Consulting & Traffic Data, Inc. provided helpful input.