Next Generation Simulation
Federal Highway Administration Taps PATH for Key Role in Filling Gaps in Functionality and Datasets
The Federal Highway Administration (FHWA) is using the California Partners for Advanced Transit and Highways (PATH) program of the UC Berkeley Institute of Transportation Studies (BITS) as a key contributor in its Next Generation Simulation (NGSIM) program. NGSIM is a multi-year project whose major goals are to address gaps in the functionality of existing traffic simulation tools, which requires new algorithm development, and to fill the need for comprehensive datasets on vehicle movement and interactions so that the algorithms used in existing models can be properly validated.
Traffic simulation models offer significant potential for evaluating existing operating conditions on freeways and other transportation facilities, and they can also help decision-makers analyze alternative operational and management strategies. However, after 35 years of development and application, they still have shortcomings.
The objective of NGSIM is to develop behavioral algorithms in support of microscopic traffic simulation, with supporting documentation and validation data sets. All NGSIM products will be freely available to simulation model developers and the transportation community at large. Current study products are posted on the NGSIM Web site.
"This work isn’t going on in a vacuum,” says PATH Director Alexander Skabardonis, who has served as senior advisor to the NGSIM effort since its inception, and who is principal investigator for the PATH NGSIM projects to produce data sets and carry out algorithm development. The research team consists of a consortium of private consulting firms, universities and other senior advisors. Three stakeholder groups (traffic modelers, software developers, and model users) oversee the work and review the results.
Using machine vision
Datasets for validating microscopic simulation algorithms (e.g., car-following or lane-changing) consist of vehicle trajectories over extended freeway segments under various traffic conditions (from free flow to congestion). Vehicle trajectories can be extracted from video recordings manually , a very time-consuming and expensive process, or automatically, using a machine-vision algorithm.
Following a survey of existing machine-vision algorithms worldwide, the NGSIM research team selected the algorithms produced as part of ongoing research at PATH to produce a prototype data set of vehicle trajectories and to assess emerging technologies to automate the trajectory generation process.
For over a decade, PATH researchers have collected video data and developed machine-vision algorithms to detect and track vehicles from video images under the sponsorship of the California Department of Transportation (Caltrans) and the National Science Foundation (NSF). One valuable result is the Berkeley Highway Lab (BHL), a unique testbed for collecting traffic surveillance data.
The BHL video surveillance system (click for enlarged view
of I-80) consists of eight digital video cameras
with overlapping fields of view on the
roof
of
a 30-story building
overlooking a section of the I-80 freeway in the City of Emeryville,
California. BHL is currently maintained by California
Center for Innovative Transportation (CCIT).
"As part of the NGSIM, we tested the existing PATH algorithms and developed a new approach, where vehicles are first detected based on their appearance then tracked by a separate tracking algorithm," Skabardonis explained. "A new vehicle detection algorithm was developed that requires significantly smaller computation."
Vehicle detection is accomplished by extracting line segments from images and matching them to 3-D vehicle models. "This algorithm shows superior vehicle localization performance in varying illumination conditions compared to the previous motion-based approaches," he added.
Once a vehicle is detected, it can be tracked based on its appearance in the camera image. A sequence of cameras is used to track the detected vehicles along a stretch of freeway. Vehicles in the overlapping area between two cameras are passed from one camera to the next by comparing their appearance in both cameras.
The algorithm can automatically generate trajectories for about 90 percent of the vehicles. To generate a complete set of trajectories from video data, researchers developed an interactive vehicle tracking system based on this algorithm (known as "caltrack" software), which runs on desktop PCs. In this system, a human operator supervises the vehicle detection and tracking procedures and confirms or modifies the results to ensure a complete set of trajectories.
The resulting system is unique; only a decades-old manual trajectory extraction system is known, Skabardonis explained. It reduces costs by 93 percent, according to a series of cost-testing analyses. "While the manual system required $250,000 to process an hour-long video of six lanes of a mile-long freeway segment, our system only required $18,000," Skabardonis said.
A video dataset collected from the BHL was used to generate a prototype dataset. The prototype dataset consists of trajectories of 4,733 vehicles over 2,952 ft. (approximately 1 km.) at 1/15 of a second, for a total of 2.8M data points. This is the largest and most comprehensive dataset on vehicle trajectories ever produced. Because it is freely available on the NGSIM site, it can be used by anyone with Internet access. Already more than 250 users have downloaded the data from the NGSIM site for various applications, Skabardonis reported.
PATH is expected to be awarded a project to develop an improved algorithm for simulating freeway flow under stop and go traffic conditions as part of NGSIM. Existing simulation models have difficulty in accurately modeling oversaturated traffic conditions on freeways. Examples include repeated stops and starts across lanes, increased lane changes to position on “moving” lanes (as perceived by drivers) or in the presence of tall vehicles (trucks, busses, SUVs), and large vehicle headways (reduced capacities).
The proposed oversaturated flow algorithm builds upon previous and ongoing research by UC Berkeley Institute of Transportation Studies (BITS) faculty and researchers supported by PATH that has produced important contributions in both theory and experimental observations. “As in the case of the data processing, this project shows the leveraging of our work and an opportunity to show people the work we do at PATH at a national level,” Skabardonis said.
Success could mean a relationship with the NGSIM project that could last for five more years and develop exciting new modeling tools, he added. “All we have to do is come up with a cool algorithm tested with real-world data in the next few months. And I think we will.”
—David Downs
Other Stories:
© 2005 UC Regents
Last Updated June 12, 2006