Why Slammed the Brakes On? Auto-annotating Driving Behaviors from Adaptive Causal Modeling

Annotating on-road driving behavior is essential for various applications like app-cabs, behavior-based insurance, etc. However, existing platforms for driving behavior analysis mainly focus on the impacts of different on-road maneuvers taken by a driver. This paper establishes that analyzing the causes behind the driving maneuvers is essential to characterize the driving behavior correctly. Accordingly, we develop DriveIn, a smartphone-based pervasive sensing system that explores both causes and impacts of driving maneuvers from video, GPS, and inertial sensor data to score a driver based on a thorough understanding of their on-road driving behavior. We evaluate DriveIn with datasets for three countries and observe that a score obtained by considering such causality factors provides a better insight into the driving behavior compared to the naive approaches.