At the Intersection of Data and Drivers

Every driver has that moment of indecision at an intersection: Is that car full of teenagers going to stop for yellow? Will it speed up? Is it safe to turn left?

In 2007, the United States Department of Transportation found almost 45 percent of all injury crashes and 22 percent of all fatal crashes took place at intersections. In most cases, the causes were attributed to drivers thinking the oncoming car would stop, obstructed views, illegal maneuvers, distraction and misjudgement of speed.

Research work coming out of The Ohio State University Department of Electrical and Computer Engineering (ECE) is now looking at the issue with the hope of saving more lives in the future.

Engineers took to the streets of Columbus this year to test out a new concept: Why not put the vehicle in charge of intersection decisions instead of the driver?

ECE Intelligent Transportation Systems (ITS) researcher Arda Kurt recently co-authored the project with colleagues from North Carolina A&T State University, in which they study how advanced driver assistance systems may help safely resolve complex road situations. The goal of the research is to develop driver behavior models to make accurate predictions for different reactions during pre-crash scenarios.

For example, when a driver decides to “Turn Left” at an intersection, key detectable actions occur first: the left turn signal blinks, the speed reduces and the brake lights illuminate. By collecting such data, algorithms can help predict potential outcomes.

As it turns out, those predictions are pretty spot-on.

According to their research, data collected to date shows an accuracy rate of over 97 percent in estimating driver intentions when approaching intersections.

“Developing precise driver behavior models near intersections can considerably reduce the number of accidents,” the research states. “The objective is to estimate from a set of observations whether the driver will stop, turn right, turn left, or go straight safely, according to traffic signal indicators.”

The concept was put to test in the streets of Columbus using driver data collected from a sensor-equipped 2012 Honda Accord provided by Ohio State.

Specifically, the data models simulate driver perception, attention, cognition and control behavior by applying different mathematical or symbolic methods.

Kurt and his fellow engineers base their research on the Hybrid-State System (HSS) framework, meaning the decisions of the driver are represented as a “discrete-state system” and the vehicle dynamics are represented as a “continuous-state system.” The proposed modeling technique utilizes the continuous observations from the vehicle and estimates the driver’s intentions at each time step using a multi-class support vector machine (SVM) approach.

SVM has already been applied to other fields of research such as handwritten recognition, face detection, text categorization, speech recognition and more.

Ultimately, through the use of on-board sensors, GPS and vehicle-to-vehicle radar, the vehicle can provide an alert to the driver, and then propose a correct action scenario of when to turn and when to allow the other vehicle to pass safely.

In addition to the intersection scenario, the proposed method could be applied to drivers merging and entering/exiting ramps, changing lanes and other high-crash risk events using large-scale data.

The research performed by Ohio State and North Carolina A&T State University was partially supported by the U.S. Department of Transportation, and the Research and Innovative Technology Administration under the University Transportation Center Program. The authors also extended their appreciation to ECE’s Umit Ozguner for his helpful insight.