News from Berkeley ITS, NewsBITS Winter 2005: X Marks the Spot, many disciplines work for intersection safety

The ITS Berkeley Online Magazine Winter 2005: Volume 1, Number 2

no left turn automated sign X Marks the Spot

Many Disciplines Team Together for Intersection Safety

The sedan pulls up to a red light in Richmond, CA, near the offices and experimental facilities of the UC Berkeley Institute of Transportation Studies' Partners for Advanced Transit and Highways (PATH) program, and the driver waits for a green light in order to make a left turn. Because there is no arrow, the maneuver will be solely up to the driver.

This is among the most dangerous situations on the road: roughly 25 percent of all crashes happen at intersections and another 20 percent are "intersection-related." Crashes are twice as likely to occur at intersections where there is a signal compared to those with no controls (no signs nor signals) at all.

The statistics argue strongly for some kind of system beyond traditional signal lights and arrows that will support the driver's decision when to execute the maneuver, which is the basis for the Intersection Decision Support (IDS) project, launched in 2002 by the Federal Highway Administration and Caltrans. Researchers affiliated with ITS Berkeley, PATH, and the Traffic Safety Center (TSC) have been working over the last five years to make situations like these safer by developing intelligent warning and detection systems that could aid drivers in these circumstances.

The undertaking has brought together engineers, psychologists, epidemiologists, computer scientists, modelers, systems designers, and communications experts, among others, to provide the technical underpinning for future intersection safety improvements.

Already, at the Richmond Field Station, not far from the real-world intersection where the sedan's driver waited to turn left, researchers have built an instrumented intersection and automated alert system prototype for the next phase of development, part of a broader federal effort to develop networks of communications between vehicles and transportation infrastructure. These efforts are expected to carry on through 2011 and possibly beyond.

Intersection Crashes Have Their Own Taxonomy

While the hazardousness of intersections makes intuitive sense, given the increased likelihood of conflicts between vehicles wherever they have the potential to cross paths, much remains to be understood about the finer details of these crashes. Traffic Safety Center researchers undertook a closer examination of the phenomenon and created a more detailed taxonomy of these crashes, building on earlier work. That enabled them to match up details like type of intersection, type of crash, nature of the signals or signage, posted speed limit, and age and gender of the driver. They then analyzed the data to guide suggestions for potential countermeasures using intelligent technologies.

The high crash rates at traditional red-yellow-green (three-phase) signal intersections suggested that an alternative would be a system that "could provide information to drivers when risk is high, but it would allow optimum traffic flow at all other times," according to TSC Director David Ragland and an author on a number of studies related to this complex problem.

Sensors monitoring oncoming and lateral traffic could activate a warning for the driver of the left-turning vehicle if they determined that a left turn was not safe to execute at that time. The problem is complex.

In the left-turn maneuver, a number of outcomes are possible:

the sedan could start on a left turn path and encounter a pedestrian in the crosswalk, which could lead it to hit the person on foot,
the sedan could be forced to slow down or stop, which could lead to a crash with oncoming traffic,
the crosswalk could be clear, but the sedan's driver could misjudge the amount of time and space available in the face of oncoming traffic and start out on the left turn path and crash into approaching traffic,
the turning driver could delay the maneuver until after the light had turned red and crash into traffic approaching laterally.

The consequences of intersection crashes tend to be more severe since 75 percent of them involve vehicles that are crossing paths, not sideswiping or rear-ending each other. Ragland stated that they killed 9,000 people a year and injured 1.5 million.

Learning How a Driver Thinks

To begin to understand the type of warning device that would work well, researchers needed to understand how drivers behave at such intersections.

How far ahead does the driver of the sedan look to see if there is a potential conflict?
What are the visual cues that the driver of the sedan uses to determine how fast approaching traffic is moving toward the car?
Are there elements such as roadway lighting, or shape or size of any approaching vehicles that might make it harder for the driver to anticipate conflicts?
If a driver does judge the speed of the approaching traffic correctly, how does the driver determine the amount of time needed to clear the intersection safely?
How is this "gap acceptance" affected by drivers' personal attributes such as age, experience, perceptiveness, and so on?
How can a system incorporate these elements into some kind of warning or perhaps guidance device?

Keeping an Eye on the Eyes on the Road: 350 Left Turns

To discover more about cues to driver behavior, PATH researchers devised instruments that were mounted inside a car to observe test subjects as they executed left turns at intersections like the one in Richmond.

Researchers Delphine Cody and Christopher Nowakowski did much of the work on these "human factors" observations and analysis, which combine psychology with engineering. Cameras recorded actions as minute as the changes in focus from one moment to the next in the driver's eyes. In one study, nine drivers made more than 350 left turns in the instrumented car that were recorded for later analysis. This data was used to quantify factors involved how a driver approaches an intersection and to develop a timeline of a crossing. The car's operating systems were monitored at the same time, so connections between braking, signaling, acceleration, and the like could be determined. A driver's pattern of acceleration proved to be a significant factor in indicating whether a driver intended to make a turn or wait.

Optometry researchers Theodore Cohn and Daniel Greenhouse expanded on their work on visual perception. They incorporated previous findings that have determined that perception is affected by the type of light, and that certain light signals travel more rapidly to the brain, enabling more time to avoid conflicts.

They have also studied the complex associations and calculations human observers make when trying to make sense of the range and velocity of approaching objects. For example, a landmark effort showed that lights that are illuminated in a sequence literally are seen more rapidly because they travel a shorter path to the brain and are processed more rapidly. Optometry research also contributed to the design of the LED pattern for the alert sign that was ultimately tested at the Richmond Field Station testbed.

Use Only When Needed

Of course, a warning system must not only be understandable and usable by the driver, it must also be accurate and timely. A major concern is fear that false positives will cause drivers to eventually disregard signs, or false negatives will cause drivers to proceed in error and risk a crash. That is where the sensors come in. A key distinctive element of IDS is the fact that, unlike traditional traffic signals, these alerts are only activated when they are needed, in order to permit unimpeded traffic flow in low-traffic and low-danger settings.

PATH engineers began by examining the usefulness of commercially available devices, in the hope that something ready-made could be deployed more quickly. They began by evaluating sensors at the Richmond Field Station, creating a testbed which was overseen by PATH engineers Ashkan Sharafsaleh and David Marco They created the prototype IDS intersection using infrared, inductive loop, radar, millimeter wave radar and other detection devices capable of determining the presence and speed of vehicles. PATH software team leader Sue Dickey enabled the intersection processor to work with these devices, and several PATH researchers—Xiao-Yun Lu, Ashkan Sharafsaleh, and Christopher Nowakowski—conducted research. The studies ascertained the timing necessary to warn driver and showed that current sensors were not adequate for such a safety-critical use.

All of this complexity needs to be reproducible in mathematical terms, in the form of a model, to allow detailed theoretical testing, study and validation of various systems. PATH researchers Steven Shladover, TSC director Ragland, PATH Researcher Xiqin Wang, and PATH engineer Joel VanderWerf teamed to design algorithms for various aspects of the models, and Joel VanderWerf built the computer model.

Determining alert criteria was done in part by Shladover, who found that providing vehicle location and speed far from the intersection were two of the more important alerts drivers would need.

What the Future Holds

This year, as the IDS project’s first phase concludes at PATH, several parallel efforts are continuing on into 2011 and possibly beyond. One component of the US DOT’s Cooperative Intersection Collision Avoidance System (CICAS) project begins in early 2006, and it is a five-year program to develop a left turn-assist sign at intersections that will tell drivers when it is unsafe to make a left turn across oncoming traffic. The project will involve a pilot test of a more evolved version of the prototype intersection at the Richmond Field Station testbed.

Because it uses communication between infrastructure and vehicles, the collision avoidance project will have a place in the Vehicle Infrastructure Integration (VII) California project, funded through the Caltrans and the US DOT, for which PATH has recently completed the first year of a three-year, $1.7 million contract. One of the key technologies is direct short range communications (DSRC) systems linking vehicles and infrastructure, and by extension an entire region’s traffic on a virtual grid of congestion, speeds, and other information to boost capacity and safety.

Engineering faculty member Raja Sengupta, who is also a researcher at the UC Berkeley Center for Future Urban Transport, A Volvo Center of Excellence, has worked extensively on DSRC architecture and functionality. One element of this network system is the “state map," which would contain all the pertinent details about an intersection’s state and be available to cars and other intersections.

“What makes this a little bit of an art is you have to bring on other disciplines to measure and understand really what the interaction is, and it’s a lot different than what you would think," explained Jim Misener, PATH Transportation Safety Research Program Lead and head of PATH's CICAS and VII California efforts. "We have a basic understanding of how everything works as drivers go through intersections; a really nice model. We’ve involved a lot of investigators from various walks to include people from the Traffic Safety Center, some computer vision people, people who understand pedestrians, to people who understand kinematics to understand how a human interacts with cars,” Misener says. “The fact is, this is soup to nuts.”

Some Helpful Links for Further IDS Inquiries:
(more recent nearer the top)

Intersection Decision Support Project Seeks to Prevent Broadside Crashes, by James A. Misener, PATH Traffic Safety Program Leader, an overview of the project. (128 K PDF).

California Intersection Decision Support: A Systems Approach to Achieve Nationally Interoperable Solutions, PATH Research Report by Ching-Yao Chan. (UCB-ITS-PRR-2005-11). (6.1 MB PDF).

Experimental Evaluation of Commercial Off-the-Shelf Sensors, by Ashkan Sharafsaleh and Ching-Yao Chan. Presented at the 2005 ITS World Congress in San Francisco. (1 MB PDF).

Effects of Traffic Density on Communication Requirements for Cooperative Intersection Collision Avoidance Systems (CICAS), by Steven Shladover, PATH. Presented at the 2005 ITS World Congress in San Francisco. (228 K PDF)

Cooperative Collision Warning: Enabling Crash Avoidance with Wireless Technology, by James A. Misener, Raja Sengupta, PATH, and Hariharan Krishnan General Motors Research and Development. Presented at the 2005 ITS World Congress in San Francisco. (600 K PDF).

Intersection Decision Support Project: Taxonomy of Crossing-Path Crashes at Intersections Using GES 2000 Data, by David Ragland and A. Zabyshny, U.C. Berkeley Traffic Safety Center. (link to eScholarship repository).

Impact of Pedestrian Presence on Movement of Left-Turning Vehicles: Method, Preliminary Results & Possible Use in Intersection Decision Support, by Ipsita Banerjee, Steven Shladover, James Misener, Ching-Yao Chan, and David Ragland, U.C. Berkeley Traffic Safety Center (link to eScholarship repository).

False Alarms and Human-Machine Warning Systems, by David Ragland and A. Zabyshny, U.C. Berkeley Traffic Safety Center. (link to eScholarship repository).

A timeline of progress on IDS, VII, and CICAS projects newsletters from PATH: http://www.path.berkeley.edu/PATH/Intellimotion/

"IDS demonstrated successfully in Washington DC’s FHWA Turner-Fairbank Research Center", story from PATH Web site.