Classification of Lateral Driving Events Using Vehicle Response Signals for Profiling Driving Behaviors

Abstract

Classification of driving events is crucial in profiling driving behaviors, which could significantly increase road safety. In several studies classifying driving events, drivers were asked to perform hard turns and lane changes. However, these represented simulated situations much different from real-life scenarios. The main aim of this study was to classify lateral driving events (turn) and non-events from naturalistic driving data in actual driving conditions. A stretch of 8 km state highway was identified as the study road, and the data were collected using 8 drivers. The acceleration and gyroscope data were collected using Inertial Measurement Unit (IMU) sensors with a frequency of approximately 200 Hz with the timestamps. A dashboard camera was fixed to capture the driver’s view with timestamps. The start and end times of the turns (left turn and right turn) and non-events were manually marked using the timestamps in the recorded videos. The total count of marked events and non-events was 1246, and their start and end times were used to label the driving events in the IMU sensor data. These labeled driving events were split into the train (934 driving events) and test (312 driving events) datasets. The Hidden Markov Model (HMM) algorithm was adopted to create classification models for the driving events. HMM models were developed using the training dataset for various features, such as lateral and longitudinal acceleration. The accuracy of these models was then compared to a test dataset. The models achieved 96.09% and 95.1% accuracy in classifying turns and non-events using data from the gyroscope’s y-axis. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.