Cooperative Localization of Elusive Dangerous Drivers

Dangerous driving is one of the primary sources of accidents worldwide. Most, if not all, research on detecting dangerous drivers has focused on collecting data from the dangerous driver and classifying their driving behavior. This approach has significant drawbacks if we cannot collect data from the driver in question. Especially when dealing with an elusive dangerous driver, he or she may disable any data collection to avoid detection.

We ask the question whether we can detect dangerous driving behavior by collecting data from nearby cooperating vehicles. A dangerous driving maneuver is likely to elicit a reaction from nearby drivers as they move out of the way to avoid the reckless driver. By pooling data from the nearby drivers, we can infer that their behavior is best explained by the presence of an elusive dangerous driver, from whom we may be missing data.

We start by collecting trajectories from cooperating vehicles. We convert the continuous trajectory information to a discrete graph representation of open space on the road segment. Next we simulate feasible trajectories of a dangerous driver through the open space graph and find likely paths, based on vehicle driving models, which strongly suggest the presence of an elusive dangerous driver missing from our dataset.

Faculty

• Kang G. Shin

Graduate Students

• Arun Ganesan