The
N T C To d a y Fall 2021
The Morgan State University National Center for Transportation Management, Research & Development Newsletter
1
Director’s Message
3
Meeting with Secretary Pete
4
The COVID-19 Dashboard
5 Research
7
Spotlight
D
Machine Learning and Distracted Driving
istracted driving is a well-publicized problem, but identifying how and when a driver is distracted requires more precise science. NTC researchers have developed a model that uses machine learning to recognize whether drivers are distracted and determine the type of distraction – i.e., texting, a handheld or handsfree phone call, voice command, or putting on and taking off clothing or eating and drinking – with respect to the different types of roads and driving environments. “This research project, A Machine Learning Model for Driving Distraction Detection, built on a previous distracted driving project conducted here using our full-size driving simulator,” Dr. Mansoureh Jeihani, the principle investigator and director of the NTC, said. “Participants drove on several types of simulated roads while researchers distracted them in different ways.” The research has resulted in several publications and a utility patent. Both projects were funded by the Maryland Department of Transportation Maryland Highway Safety Office. Machine learning is a term that refers to computer programs that can learn from data, make a prediction, and classify a data set. Machine learning algorithms have been used in utilities management, vehicle-to-cloud communications, and health care. “Distraction is a complex problem because it occurs in three ways, and in combinations of those ways, all of which can contribute to a crash, and it can be deceiving,” Dr. Jeihani noted. Visual distraction – a billboard or something on the side of the road – takes a driver’s eyes off the road. Manual distractions occur when a driver takes their hands off the wheel, to unwrap a burger or take off a jacket. Cognitive distraction is when a driver thinks about something else.
Some distractions, like sending a text message, involve all three. To further complicate matters, a driver who is engaged in a distracting task is not always distracted, and one with both hands on the steering wheel might be lost in thought. Thanks to advances in technology, automotive manufacturers can implement more safety features, resulting in automated vehicles with applications that use a combination of data and hardware such as cameras and lidar sensors (a variable distance range measuring sensor) to make the driving experience as safe as possible. “The machine learning predictions and identification developed in this study contribute to such applications and to the development of after-market products,” Dr. Jeihani said. In the previous phase of the study, the mean value of variables such as speed, braking, steering, lane changing patterns, throttle, steering velocity, and distance from the center of the road were used to identify distraction behavior in different scenarios on different roads. This time, the researchers developed new variables that could formulate the sudden changes in driving behavior to better identify distraction behaviors. The driving performance data from the previous study were transformed to show the rate of change in driving performance in every second. One second was chosen for data analysis because time is such a crucial factor in distracted driving and crash risk. The best machine learning model demonstrated 72% prediction power in a validation set to distinguish hand-held call and voice-to-text distraction types. Fairly small data sets limited this study, and using more driving simulator or real-world data could improve the model. The entire report is available at 2020 10 13 DDML_FinalReport-.pdf (morgan.edu)