Hidden Markov Model based driving event detection and driver profiling from mobile inertial sensor data

With the advent of smartphones and advancements in sensor capabilities, it is possible to actively monitor drivers and provide a viable solution necessary to reduce vehicle accidents. Driving maneuvers provide an insight to a driver's driving skills and behavior, which is an important aspect for applications such as driver profiling, driver safety, fuel consumption modeling, etc. Driver profiling requires detection of sharp and normal driving maneuvers having high and low Signal-to-Noise Ratio (SNR), respectively. Typical event detection techniques detect sharp driving maneuvers but fail to detect normal maneuvers. In this paper, we propose Hidden Markov Model (HMM) based technique to detect lateral maneuvers and Jerk Energy based technique to detect longitudinal maneuvers. Most driver profiling techniques consider only longitudinal events such as hard acceleration/braking, whereas the proposed approach profiles a driver by coupling lateral and longitudinal events. Based on collected datasets on diverse type of driving scenario, events are detected with 95% accuracy. For driver profiling, we achieve 90% accuracy in match between drivers subjective score and model-based estimated score.