Analysis and Classification of Driver Behavior using In-Vehicle CAN-Bus Information

This paper describes recent advances in the analysis and classification of driver behavior in actual driving scenarios. We employ data obtained from the UTDrive corpus to model driving behavior and to detect if distraction due to secondary tasks is present. Hidden Markov Models (HMMs) are used to capture the sequence of driving characteristics acquired from the vehicle’s CAN-Bus (Controller Area Network) information. Driver behavior is described and modeled using data from steering wheel angle, brake status, acceleration status, and vehicle speed. We evaluate data and models in three distinct classification tasks: 1) action classification, 2) distraction detection, and 3) driver identification. The aim of action classification is to categorize long-term driving behaviors such as turning, lane changing, stopping, and constant/no change (neutral driving). The goal of driver identification task is to classify drivers from their driving behavior characteristics, and distraction detection identifies whether the driver is under distraction due to secondary tasks. Experiments were conducted using 9 drivers from the UTDrive corpus. We report accuracy on modeling driver behavior based on these studies and discuss our future work. Initial results show that event detection for driving can be accomplished at rates ranging from 30-70% depending on the number of unique conditions based on CAN-Bus signals.