Learning to Associate Observed Driver Behavior with Traffic Controls

Adaptive techniques support the development of new tools to help traffic engineers classify and evaluate traffic flow at intersections. We describe a tool that learns to associate driver behavior with a subset of traffic controls, e.g. stoplights and stop signs. In the case where the traffic controls for an intersection are not readily available or are unknown, the tool automatically identifies the traffic controls present at an intersection from observed driver behavior. This capability may be used to augment digital maps with traffic control locations. In the case where traffic controls are known or have previously been classified, the tool flags instances where driver behavior is inconsistent with the traffic controls actually present. This capability might be used by various services for drivers such as dynamic routing and new safety systems. It might also be used by traffic engineers to evaluate control placement in real or simulated road networks by finding situations that elicit unusual driver behavior. We calibrate the tool with driving data on a set of segments with known controls. The tool first learns to identify controls present on individual road segments, then uses hand-crafted rules to verify control consistency across segments at intersections. The data set comprised real-world position data collected during normal daily driving. The tool accurately identified 100% of the data that passed verification. These results encourage us to believe that the system can provide traffic engineers with a reliable mapping between driver behavior and traffic controls. Christopher A. Pribe and Seth O. Rogers 1