Jump to content

Iowa State researchers use machine learning to improve road safety and efficiency

Using Google's TensorFlow, Iowa Department of Transportation teams up with researchers at Iowa State University to gain insights into traffic behavior and to support traffic operations decision making.

Covering approximately 115,000 miles, Iowa's roadway system is one of the largest (per capita) in the nation. Keeping all those roads safe for travelers throughout the year is a challenge. Car crashes tend to spike during the winter months, when the state receives an average of 33 inches of snow, causing costly delays and hazards for drivers. The Iowa Department of Transportation (DOT) plays a critical role in assessing the state of the roadways in real-time: it is their job to notify first responders and commuters about crashes and current road conditions. Getting children to and from school is a top priority every day. Mike Patton, superintendent of the Roland-Story school district, notes that "our primary concern is making sure that we get everyone to and from school safely. As road conditions change it's important for us to stay on top of what's happening out there. If technology could help us determine what's occurring across the entire school district we'd make better decisions for our kids."

As part of that goal, the Iowa DOT collects mountains of data, generating daily, monthly, and annual reports across a range of variables, from numbers of fatalities to average vehicle miles travelled (VMT). Although roads are maintained by cities and counties as well as the state, the state DOT has responsibility for all of that data collection. Bonnie Castillo, Director of the Traffic Management Center, reports that the Iowa DOT has operators continuously monitoring road conditions more than 400 cameras as well as other sensors, like radar detectors, GPS devices, and cameras mounted on snow plows. As the numbers and images accumulate they must be interpreted. So the Iowa DOT turned to Iowa State's Institute of Transportation (InTrans) for help. Since 2013 they have been collaborating through the university's REACTOR lab, one of the only facilities in the country using big data to analyze problems and recommend solutions for transportation. The lab collects terabytes of data from a wide variety of sources as frequently as every twenty seconds. "We wanted to condense that data and analyze it much more quickly," says Tracey Bramble, Information Specialist at the Iowa DOT.

TensorFlow is helping us understand the traffic flow and with that information we can be better at alerting responders and being predictive.

Neal Hawkins, Associate Director of the Institute of Transportation, Iowa State University

Using AI to make roads safer

Real-time traffic data presents a challenge that is perfect for applying high-performance machine learning, which can process numerical computations on large amounts of data flexibly and fast. Using TensorFlow, Google's open source machine-learning platform, InTrans researchers could easily mine all that raw information for new insights, like identifying patterns of road congestion and improving response times to traffic incidents. Neal Hawkins, Associate Director of InTrans, says that "TensorFlow is helping us understand the traffic flow and with that information we can be better at alerting responders and being predictive." Tingting Huang, Research Scientist at InTrans, adds that "machine learning can automatically check images from the camera to detect if a crash happens so we can send the information to traffic engineers."

The future is here

Iowa's DOT innovative approach and research partnership is just one example of how machine learning can improve transportation across the United States. In California, for example, college students used TensorFlow to identify pot holes and dangerous road cracks in Los Angeles. The team at InTrans is already working on other AI-driven transportation studies, like how drivers in different demographics respond to traffic signs, work zones, and detours. Combining behavior datasets with other variables could lead to "personalized transportation" with smart cars that could advise drivers when to take a break. Machine learning could even calculate the probability of crashes caused by different weather conditions and warn drivers in advance. "Machine learning has really revolutionized the way we think about the data we have and the connection we have to vehicles," says Hawkins. Castillo concludes that "the Iowa roadways are safer because of all the technology we've been able to bring. We're making travel more safe and efficient and I only see that getting better."

The Iowa roadways are safer because of all the technology we've been able to bring. We're making travel more safe and efficient and I only see that getting better.

Bonnie Castillo, Director of Traffic Management, Iowa Department of Transportation

Sign up here for updates, insights, resources, and more.